2 * Copyright 2000, International Business Machines Corporation and others.
5 * This software has been released under the terms of the IBM Public
6 * License. For details, see the LICENSE file in the top-level source
7 * directory or online at http://www.openafs.org/dl/license10.html
9 * Portions Copyright (c) 2005-2008 Sine Nomine Associates
12 /* 1/1/89: NB: this stuff is all going to be replaced. Don't take it too seriously */
17 Institution: The Information Technology Center, Carnegie-Mellon University
21 #include <afsconfig.h>
22 #include <afs/param.h>
30 #ifdef HAVE_SYS_FILE_H
34 #ifdef AFS_PTHREAD_ENV
35 # include <opr/lock.h>
37 # include <opr/lockstub.h>
40 #include <afs/afsint.h>
42 #include <rx/rx_queue.h>
45 #if !defined(AFS_SGI_ENV)
48 #else /* AFS_OSF_ENV */
49 #ifdef AFS_VFSINCL_ENV
52 #include <sys/fs/ufs_fs.h>
54 #if defined(AFS_DARWIN_ENV) || defined(AFS_XBSD_ENV)
55 #include <ufs/ufs/dinode.h>
56 #include <ufs/ffs/fs.h>
61 #else /* AFS_VFSINCL_ENV */
62 #if !defined(AFS_AIX_ENV) && !defined(AFS_LINUX20_ENV) && !defined(AFS_XBSD_ENV) && !defined(AFS_DARWIN_ENV)
65 #endif /* AFS_VFSINCL_ENV */
66 #endif /* AFS_OSF_ENV */
67 #endif /* AFS_SGI_ENV */
68 #endif /* !AFS_NT40_ENV */
76 #if defined(AFS_SUN_ENV) || defined(AFS_SUN5_ENV)
78 #include <sys/mnttab.h>
79 #include <sys/mntent.h>
85 #if defined(AFS_SGI_ENV)
88 #ifndef AFS_LINUX20_ENV
89 #include <fstab.h> /* Need to find in libc 5, present in libc 6 */
92 #endif /* AFS_SGI_ENV */
94 #endif /* AFS_HPUX_ENV */
98 #include <afs/errors.h>
101 #include <afs/afssyscalls.h>
103 #include <afs/afsutil.h>
104 #include "daemon_com.h"
106 #include "salvsync.h"
109 #include "partition.h"
110 #include "volume_inline.h"
115 #ifdef AFS_PTHREAD_ENV
116 pthread_mutex_t vol_glock_mutex;
117 pthread_mutex_t vol_trans_mutex;
118 pthread_cond_t vol_put_volume_cond;
119 pthread_cond_t vol_sleep_cond;
120 pthread_cond_t vol_init_attach_cond;
121 pthread_cond_t vol_vinit_cond;
122 int vol_attach_threads = 1;
123 #endif /* AFS_PTHREAD_ENV */
125 /* start-time configurable I/O parameters */
126 ih_init_params vol_io_params;
128 #ifdef AFS_DEMAND_ATTACH_FS
129 pthread_mutex_t vol_salvsync_mutex;
132 * Set this to 1 to disallow SALVSYNC communication in all threads; used
133 * during shutdown, since the salvageserver may have gone away.
135 static volatile sig_atomic_t vol_disallow_salvsync = 0;
136 #endif /* AFS_DEMAND_ATTACH_FS */
139 * has VShutdown_r been called / is VShutdown_r running?
141 static int vol_shutting_down = 0;
144 extern void *calloc(), *realloc();
147 /* Forward declarations */
148 static Volume *attach2(Error * ec, VolumeId volumeId, char *path,
149 struct DiskPartition64 *partp, Volume * vp,
150 int isbusy, int mode, int *acheckedOut);
151 static void ReallyFreeVolume(Volume * vp);
152 #ifdef AFS_DEMAND_ATTACH_FS
153 static void FreeVolume(Volume * vp);
154 #else /* !AFS_DEMAND_ATTACH_FS */
155 #define FreeVolume(vp) ReallyFreeVolume(vp)
156 static void VScanUpdateList(void);
157 #endif /* !AFS_DEMAND_ATTACH_FS */
158 static void VInitVolumeHeaderCache(afs_uint32 howMany);
159 static int GetVolumeHeader(Volume * vp);
160 static void ReleaseVolumeHeader(struct volHeader *hd);
161 static void FreeVolumeHeader(Volume * vp);
162 static void AddVolumeToHashTable(Volume * vp, VolumeId hashid);
163 static void DeleteVolumeFromHashTable(Volume * vp);
165 static int VHold(Volume * vp);
167 static int VHold_r(Volume * vp);
168 static void VGetBitmap_r(Error * ec, Volume * vp, VnodeClass class);
169 static void VReleaseVolumeHandles_r(Volume * vp);
170 static void VCloseVolumeHandles_r(Volume * vp);
171 static void LoadVolumeHeader(Error * ec, Volume * vp);
172 static int VCheckOffline(Volume * vp);
173 static int VCheckDetach(Volume * vp);
174 static Volume * GetVolume(Error * ec, Error * client_ec, VolumeId volumeId,
175 Volume * hint, const struct timespec *ts);
177 int LogLevel; /* Vice loglevel--not defined as extern so that it will be
178 * defined when not linked with vice, XXXX */
179 ProgramType programType; /* The type of program using the package */
180 static VolumePackageOptions vol_opts;
182 /* extended volume package statistics */
185 #ifdef VOL_LOCK_DEBUG
186 pthread_t vol_glock_holder = 0;
190 /* this parameter needs to be tunable at runtime.
191 * 128 was really inadequate for largish servers -- at 16384 volumes this
192 * puts average chain length at 128, thus an average 65 deref's to find a volptr.
193 * talk about bad spatial locality...
195 * an AVL or splay tree might work a lot better, but we'll just increase
196 * the default hash table size for now
198 #define DEFAULT_VOLUME_HASH_SIZE 256 /* Must be a power of 2!! */
199 #define DEFAULT_VOLUME_HASH_MASK (DEFAULT_VOLUME_HASH_SIZE-1)
200 #define VOLUME_HASH(volumeId) (volumeId&(VolumeHashTable.Mask))
203 * turn volume hash chains into partially ordered lists.
204 * when the threshold is exceeded between two adjacent elements,
205 * perform a chain rebalancing operation.
207 * keep the threshold high in order to keep cache line invalidates
208 * low "enough" on SMPs
210 #define VOLUME_HASH_REORDER_THRESHOLD 200
213 * when possible, don't just reorder single elements, but reorder
214 * entire chains of elements at once. a chain of elements that
215 * exceed the element previous to the pivot by at least CHAIN_THRESH
216 * accesses are moved in front of the chain whose elements have at
217 * least CHAIN_THRESH less accesses than the pivot element
219 #define VOLUME_HASH_REORDER_CHAIN_THRESH (VOLUME_HASH_REORDER_THRESHOLD / 2)
221 #include "rx/rx_queue.h"
224 VolumeHashTable_t VolumeHashTable = {
225 DEFAULT_VOLUME_HASH_SIZE,
226 DEFAULT_VOLUME_HASH_MASK,
231 static void VInitVolumeHash(void);
235 /* This macro is used where an ffs() call does not exist. Was in util/ffs.c */
239 afs_int32 ffs_tmp = x;
243 for (ffs_i = 1;; ffs_i++) {
250 #endif /* !AFS_HAVE_FFS */
252 #ifdef AFS_PTHREAD_ENV
254 * disk partition queue element
256 typedef struct diskpartition_queue_t {
257 struct rx_queue queue; /**< queue header */
258 struct DiskPartition64 *diskP; /**< disk partition table entry */
259 } diskpartition_queue_t;
261 #ifndef AFS_DEMAND_ATTACH_FS
263 typedef struct vinitvolumepackage_thread_t {
264 struct rx_queue queue;
265 pthread_cond_t thread_done_cv;
266 int n_threads_complete;
267 } vinitvolumepackage_thread_t;
268 static void * VInitVolumePackageThread(void * args);
270 #else /* !AFS_DEMAND_ATTTACH_FS */
271 #define VINIT_BATCH_MAX_SIZE 512
274 * disk partition work queue
276 struct partition_queue {
277 struct rx_queue head; /**< diskpartition_queue_t queue */
278 pthread_mutex_t mutex;
283 * volumes parameters for preattach
285 struct volume_init_batch {
286 struct rx_queue queue; /**< queue header */
287 int thread; /**< posting worker thread */
288 int last; /**< indicates thread is done */
289 int size; /**< number of volume ids in batch */
290 Volume *batch[VINIT_BATCH_MAX_SIZE]; /**< volumes ids to preattach */
294 * volume parameters work queue
296 struct volume_init_queue {
297 struct rx_queue head; /**< volume_init_batch queue */
298 pthread_mutex_t mutex;
303 * volume init worker thread parameters
305 struct vinitvolumepackage_thread_param {
306 int nthreads; /**< total number of worker threads */
307 int thread; /**< thread number for this worker thread */
308 struct partition_queue *pq; /**< queue partitions to scan */
309 struct volume_init_queue *vq; /**< queue of volume to preattach */
312 static void *VInitVolumePackageThread(void *args);
313 static struct DiskPartition64 *VInitNextPartition(struct partition_queue *pq);
314 static VolumeId VInitNextVolumeId(DIR *dirp);
315 static int VInitPreAttachVolumes(int nthreads, struct volume_init_queue *vq);
317 #endif /* !AFS_DEMAND_ATTACH_FS */
318 #endif /* AFS_PTHREAD_ENV */
320 #ifndef AFS_DEMAND_ATTACH_FS
321 static int VAttachVolumesByPartition(struct DiskPartition64 *diskP,
322 int * nAttached, int * nUnattached);
323 #endif /* AFS_DEMAND_ATTACH_FS */
326 #ifdef AFS_DEMAND_ATTACH_FS
327 /* demand attach fileserver extensions */
330 * in the future we will support serialization of VLRU state into the fs_state
333 * these structures are the beginning of that effort
335 struct VLRU_DiskHeader {
336 struct versionStamp stamp; /* magic and structure version number */
337 afs_uint32 mtime; /* time of dump to disk */
338 afs_uint32 num_records; /* number of VLRU_DiskEntry records */
341 struct VLRU_DiskEntry {
342 VolumeId vid; /* volume ID */
343 afs_uint32 idx; /* generation */
344 afs_uint32 last_get; /* timestamp of last get */
347 struct VLRU_StartupQueue {
348 struct VLRU_DiskEntry * entry;
353 typedef struct vshutdown_thread_t {
355 pthread_mutex_t lock;
357 pthread_cond_t master_cv;
359 int n_threads_complete;
361 int schedule_version;
364 byte n_parts_done_pass;
365 byte part_thread_target[VOLMAXPARTS+1];
366 byte part_done_pass[VOLMAXPARTS+1];
367 struct rx_queue * part_pass_head[VOLMAXPARTS+1];
368 int stats[4][VOLMAXPARTS+1];
369 } vshutdown_thread_t;
370 static void * VShutdownThread(void * args);
373 static Volume * VAttachVolumeByVp_r(Error * ec, Volume * vp, int mode);
374 static int VCheckFree(Volume * vp);
377 static void AddVolumeToVByPList_r(Volume * vp);
378 static void DeleteVolumeFromVByPList_r(Volume * vp);
379 static void VVByPListBeginExclusive_r(struct DiskPartition64 * dp);
380 static void VVByPListEndExclusive_r(struct DiskPartition64 * dp);
381 static void VVByPListWait_r(struct DiskPartition64 * dp);
383 /* online salvager */
385 VCHECK_SALVAGE_OK = 0, /**< no pending salvage */
386 VCHECK_SALVAGE_SCHEDULED = 1, /**< salvage has been scheduled */
387 VCHECK_SALVAGE_ASYNC = 2, /**< salvage being scheduled */
388 VCHECK_SALVAGE_DENIED = 3, /**< salvage not scheduled; denied */
389 VCHECK_SALVAGE_FAIL = 4 /**< salvage not scheduled; failed */
391 static int VCheckSalvage(Volume * vp);
392 #if defined(SALVSYNC_BUILD_CLIENT) || defined(FSSYNC_BUILD_CLIENT)
393 static int VScheduleSalvage_r(Volume * vp);
396 /* Volume hash table */
397 static void VReorderHash_r(VolumeHashChainHead * head, Volume * pp, Volume * vp);
398 static void VHashBeginExclusive_r(VolumeHashChainHead * head);
399 static void VHashEndExclusive_r(VolumeHashChainHead * head);
400 static void VHashWait_r(VolumeHashChainHead * head);
403 static int ShutdownVByPForPass_r(struct DiskPartition64 * dp, int pass);
404 static int ShutdownVolumeWalk_r(struct DiskPartition64 * dp, int pass,
405 struct rx_queue ** idx);
406 static void ShutdownController(vshutdown_thread_t * params);
407 static void ShutdownCreateSchedule(vshutdown_thread_t * params);
410 static void VLRU_ComputeConstants(void);
411 static void VInitVLRU(void);
412 static void VLRU_Init_Node_r(Volume * vp);
413 static void VLRU_Add_r(Volume * vp);
414 static void VLRU_Delete_r(Volume * vp);
415 static void VLRU_UpdateAccess_r(Volume * vp);
416 static void * VLRU_ScannerThread(void * args);
417 static void VLRU_Scan_r(int idx);
418 static void VLRU_Promote_r(int idx);
419 static void VLRU_Demote_r(int idx);
420 static void VLRU_SwitchQueues(Volume * vp, int new_idx, int append);
423 static int VCheckSoftDetach(Volume * vp, afs_uint32 thresh);
424 static int VCheckSoftDetachCandidate(Volume * vp, afs_uint32 thresh);
425 static int VSoftDetachVolume_r(Volume * vp, afs_uint32 thresh);
428 pthread_key_t VThread_key;
429 VThreadOptions_t VThread_defaults = {
430 0 /**< allow salvsync */
432 #endif /* AFS_DEMAND_ATTACH_FS */
435 struct Lock vol_listLock; /* Lock obtained when listing volumes:
436 * prevents a volume from being missed
437 * if the volume is attached during a
441 /* Common message used when the volume goes off line */
442 char *VSalvageMessage =
443 "Files in this volume are currently unavailable; call operations";
445 int VInit; /* 0 - uninitialized,
446 * 1 - initialized but not all volumes have been attached,
447 * 2 - initialized and all volumes have been attached,
448 * 3 - initialized, all volumes have been attached, and
449 * VConnectFS() has completed. */
451 static int vinit_attach_abort = 0;
453 bit32 VolumeCacheCheck; /* Incremented everytime a volume goes on line--
454 * used to stamp volume headers and in-core
455 * vnodes. When the volume goes on-line the
456 * vnode will be invalidated
457 * access only with VOL_LOCK held */
462 /***************************************************/
463 /* Startup routines */
464 /***************************************************/
466 #if defined(FAST_RESTART) && defined(AFS_DEMAND_ATTACH_FS)
467 # error FAST_RESTART and DAFS are incompatible. For the DAFS equivalent \
468 of FAST_RESTART, use the -unsafe-nosalvage fileserver argument
472 * assign default values to a VolumePackageOptions struct.
474 * Always call this on a VolumePackageOptions struct first, then set any
475 * specific options you want, then call VInitVolumePackage2.
477 * @param[in] pt caller's program type
478 * @param[out] opts volume package options
481 VOptDefaults(ProgramType pt, VolumePackageOptions *opts)
483 opts->nLargeVnodes = opts->nSmallVnodes = 5;
486 opts->canScheduleSalvage = 0;
487 opts->canUseFSSYNC = 0;
488 opts->canUseSALVSYNC = 0;
490 opts->interrupt_rxcall = NULL;
491 opts->offline_timeout = -1;
492 opts->offline_shutdown_timeout = -1;
493 opts->usage_threshold = 128;
494 opts->usage_rate_limit = 5;
497 opts->unsafe_attach = 1;
498 #else /* !FAST_RESTART */
499 opts->unsafe_attach = 0;
500 #endif /* !FAST_RESTART */
504 opts->canScheduleSalvage = 1;
505 opts->canUseSALVSYNC = 1;
509 opts->canUseFSSYNC = 1;
513 opts->nLargeVnodes = 0;
514 opts->nSmallVnodes = 0;
516 opts->canScheduleSalvage = 1;
517 opts->canUseFSSYNC = 1;
527 * Set VInit to a certain value, and signal waiters.
529 * @param[in] value the value to set VInit to
534 VSetVInit_r(int value)
537 opr_cv_broadcast(&vol_vinit_cond);
541 VLogOfflineTimeout(const char *type, afs_int32 timeout)
547 Log("VInitVolumePackage: Interrupting clients accessing %s "
548 "immediately\n", type);
550 Log("VInitVolumePackage: Interrupting clients accessing %s "
551 "after %ld second%s\n", type, (long)timeout, timeout==1?"":"s");
556 VInitVolumePackage2(ProgramType pt, VolumePackageOptions * opts)
558 int errors = 0; /* Number of errors while finding vice partitions. */
563 #ifndef AFS_PTHREAD_ENV
564 if (opts->offline_timeout != -1 || opts->offline_shutdown_timeout != -1) {
565 Log("VInitVolumePackage: offline_timeout and/or "
566 "offline_shutdown_timeout was specified, but the volume package "
567 "does not support these for LWP builds\n");
571 VLogOfflineTimeout("volumes going offline", opts->offline_timeout);
572 VLogOfflineTimeout("volumes going offline during shutdown",
573 opts->offline_shutdown_timeout);
575 memset(&VStats, 0, sizeof(VStats));
576 VStats.hdr_cache_size = 200;
578 VInitPartitionPackage();
580 #ifdef AFS_DEMAND_ATTACH_FS
581 if (programType == fileServer) {
584 VLRU_SetOptions(VLRU_SET_ENABLED, 0);
586 opr_Verify(pthread_key_create(&VThread_key, NULL) == 0);
589 opr_mutex_init(&vol_glock_mutex);
590 opr_mutex_init(&vol_trans_mutex);
591 opr_cv_init(&vol_put_volume_cond);
592 opr_cv_init(&vol_sleep_cond);
593 opr_cv_init(&vol_init_attach_cond);
594 opr_cv_init(&vol_vinit_cond);
595 #ifndef AFS_PTHREAD_ENV
597 #endif /* AFS_PTHREAD_ENV */
598 Lock_Init(&vol_listLock);
600 srandom(time(0)); /* For VGetVolumeInfo */
602 #ifdef AFS_DEMAND_ATTACH_FS
603 opr_mutex_init(&vol_salvsync_mutex);
604 #endif /* AFS_DEMAND_ATTACH_FS */
606 /* Ok, we have done enough initialization that fileserver can
607 * start accepting calls, even though the volumes may not be
608 * available just yet.
612 #if defined(AFS_DEMAND_ATTACH_FS) && defined(SALVSYNC_BUILD_SERVER)
613 if (programType == salvageServer) {
616 #endif /* AFS_DEMAND_ATTACH_FS */
617 #ifdef FSSYNC_BUILD_SERVER
618 if (programType == fileServer) {
622 #if defined(AFS_DEMAND_ATTACH_FS) && defined(SALVSYNC_BUILD_CLIENT)
623 if (VCanUseSALVSYNC()) {
624 /* establish a connection to the salvager at this point */
625 opr_Verify(VConnectSALV() != 0);
627 #endif /* AFS_DEMAND_ATTACH_FS */
629 if (opts->volcache > VStats.hdr_cache_size)
630 VStats.hdr_cache_size = opts->volcache;
631 VInitVolumeHeaderCache(VStats.hdr_cache_size);
633 VInitVnodes(vLarge, opts->nLargeVnodes);
634 VInitVnodes(vSmall, opts->nSmallVnodes);
637 errors = VAttachPartitions();
641 if (programType != fileServer) {
642 errors = VInitAttachVolumes(programType);
648 #ifdef FSSYNC_BUILD_CLIENT
649 if (VCanUseFSSYNC()) {
651 #ifdef AFS_DEMAND_ATTACH_FS
652 if (programType == salvageServer) {
653 Log("Unable to connect to file server; aborted\n");
656 #endif /* AFS_DEMAND_ATTACH_FS */
657 Log("Unable to connect to file server; will retry at need\n");
660 #endif /* FSSYNC_BUILD_CLIENT */
665 #if !defined(AFS_PTHREAD_ENV)
667 * Attach volumes in vice partitions
669 * @param[in] pt calling program type
672 * @note This is the original, non-threaded version of attach parititions.
674 * @post VInit state is 2
677 VInitAttachVolumes(ProgramType pt)
679 opr_Assert(VInit==1);
680 if (pt == fileServer) {
681 struct DiskPartition64 *diskP;
682 /* Attach all the volumes in this partition */
683 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
684 int nAttached = 0, nUnattached = 0;
685 opr_Verify(VAttachVolumesByPartition(diskP,
686 &nAttached, &nUnattached)
691 VSetVInit_r(2); /* Initialized, and all volumes have been attached */
692 LWP_NoYieldSignal(VInitAttachVolumes);
696 #endif /* !AFS_PTHREAD_ENV */
698 #if defined(AFS_PTHREAD_ENV) && !defined(AFS_DEMAND_ATTACH_FS)
700 * Attach volumes in vice partitions
702 * @param[in] pt calling program type
705 * @note Threaded version of attach parititions.
707 * @post VInit state is 2
710 VInitAttachVolumes(ProgramType pt)
712 opr_Assert(VInit==1);
713 if (pt == fileServer) {
714 struct DiskPartition64 *diskP;
715 struct vinitvolumepackage_thread_t params;
716 struct diskpartition_queue_t * dpq;
717 int i, threads, parts;
719 pthread_attr_t attrs;
721 opr_cv_init(¶ms.thread_done_cv);
723 params.n_threads_complete = 0;
725 /* create partition work queue */
726 for (parts=0, diskP = DiskPartitionList; diskP; diskP = diskP->next, parts++) {
727 dpq = malloc(sizeof(struct diskpartition_queue_t));
728 opr_Assert(dpq != NULL);
730 queue_Append(¶ms,dpq);
733 threads = min(parts, vol_attach_threads);
736 /* spawn off a bunch of initialization threads */
737 opr_Verify(pthread_attr_init(&attrs) == 0);
738 opr_Verify(pthread_attr_setdetachstate(&attrs,
739 PTHREAD_CREATE_DETACHED)
742 Log("VInitVolumePackage: beginning parallel fileserver startup\n");
743 Log("VInitVolumePackage: using %d threads to attach volumes on %d partitions\n",
747 for (i=0; i < threads; i++) {
750 opr_Verify(pthread_create(&tid, &attrs,
751 &VInitVolumePackageThread,
753 AFS_SIGSET_RESTORE();
756 while(params.n_threads_complete < threads) {
757 VOL_CV_WAIT(¶ms.thread_done_cv);
761 opr_Verify(pthread_attr_destroy(&attrs) == 0);
763 /* if we're only going to run one init thread, don't bother creating
765 Log("VInitVolumePackage: beginning single-threaded fileserver startup\n");
766 Log("VInitVolumePackage: using 1 thread to attach volumes on %d partition(s)\n",
769 VInitVolumePackageThread(¶ms);
772 opr_cv_destroy(¶ms.thread_done_cv);
775 VSetVInit_r(2); /* Initialized, and all volumes have been attached */
776 opr_cv_broadcast(&vol_init_attach_cond);
782 VInitVolumePackageThread(void * args) {
784 struct DiskPartition64 *diskP;
785 struct vinitvolumepackage_thread_t * params;
786 struct diskpartition_queue_t * dpq;
788 params = (vinitvolumepackage_thread_t *) args;
792 /* Attach all the volumes in this partition */
793 while (queue_IsNotEmpty(params)) {
794 int nAttached = 0, nUnattached = 0;
796 if (vinit_attach_abort) {
797 Log("Aborting initialization\n");
801 dpq = queue_First(params,diskpartition_queue_t);
807 opr_Verify(VAttachVolumesByPartition(diskP, &nAttached,
814 params->n_threads_complete++;
815 opr_cv_signal(¶ms->thread_done_cv);
819 #endif /* AFS_PTHREAD_ENV && !AFS_DEMAND_ATTACH_FS */
821 #if defined(AFS_DEMAND_ATTACH_FS)
823 * Attach volumes in vice partitions
825 * @param[in] pt calling program type
828 * @note Threaded version of attach partitions.
830 * @post VInit state is 2
833 VInitAttachVolumes(ProgramType pt)
835 opr_Assert(VInit==1);
836 if (pt == fileServer) {
838 struct DiskPartition64 *diskP;
839 struct partition_queue pq;
840 struct volume_init_queue vq;
842 int i, threads, parts;
844 pthread_attr_t attrs;
846 /* create partition work queue */
849 opr_mutex_init(&pq.mutex);
850 for (parts = 0, diskP = DiskPartitionList; diskP; diskP = diskP->next, parts++) {
851 struct diskpartition_queue_t *dp;
852 dp = malloc(sizeof(struct diskpartition_queue_t));
853 opr_Assert(dp != NULL);
855 queue_Append(&pq, dp);
858 /* number of worker threads; at least one, not to exceed the number of partitions */
859 threads = min(parts, vol_attach_threads);
861 /* create volume work queue */
864 opr_mutex_init(&vq.mutex);
866 opr_Verify(pthread_attr_init(&attrs) == 0);
867 opr_Verify(pthread_attr_setdetachstate(&attrs,
868 PTHREAD_CREATE_DETACHED) == 0);
870 Log("VInitVolumePackage: beginning parallel fileserver startup\n");
871 Log("VInitVolumePackage: using %d threads to pre-attach volumes on %d partitions\n",
874 /* create threads to scan disk partitions. */
875 for (i=0; i < threads; i++) {
876 struct vinitvolumepackage_thread_param *params;
879 params = malloc(sizeof(struct vinitvolumepackage_thread_param));
883 params->nthreads = threads;
884 params->thread = i+1;
887 opr_Verify(pthread_create(&tid, &attrs,
888 &VInitVolumePackageThread,
889 (void*)params) == 0);
890 AFS_SIGSET_RESTORE();
893 VInitPreAttachVolumes(threads, &vq);
895 opr_Verify(pthread_attr_destroy(&attrs) == 0);
896 opr_cv_destroy(&pq.cv);
897 opr_mutex_destroy(&pq.mutex);
898 opr_cv_destroy(&vq.cv);
899 opr_mutex_destroy(&vq.mutex);
903 VSetVInit_r(2); /* Initialized, and all volumes have been attached */
904 opr_cv_broadcast(&vol_init_attach_cond);
911 * Volume package initialization worker thread. Scan partitions for volume
912 * header files. Gather batches of volume ids and dispatch them to
913 * the main thread to be preattached. The volume preattachement is done
914 * in the main thread to avoid global volume lock contention.
917 VInitVolumePackageThread(void *args)
919 struct vinitvolumepackage_thread_param *params;
920 struct DiskPartition64 *partition;
921 struct partition_queue *pq;
922 struct volume_init_queue *vq;
923 struct volume_init_batch *vb;
926 params = (struct vinitvolumepackage_thread_param *)args;
932 vb = malloc(sizeof(struct volume_init_batch));
934 vb->thread = params->thread;
938 Log("Scanning partitions on thread %d of %d\n", params->thread, params->nthreads);
939 while((partition = VInitNextPartition(pq))) {
943 Log("Partition %s: pre-attaching volumes\n", partition->name);
944 dirp = opendir(VPartitionPath(partition));
946 Log("opendir on Partition %s failed, errno=%d!\n", partition->name, errno);
949 while ((vid = VInitNextVolumeId(dirp))) {
950 Volume *vp = calloc(1, sizeof(Volume));
952 vp->device = partition->device;
953 vp->partition = partition;
955 queue_Init(&vp->vnode_list);
956 queue_Init(&vp->rx_call_list);
957 opr_cv_init(&V_attachCV(vp));
959 vb->batch[vb->size++] = vp;
960 if (vb->size == VINIT_BATCH_MAX_SIZE) {
961 opr_mutex_enter(&vq->mutex);
962 queue_Append(vq, vb);
963 opr_cv_broadcast(&vq->cv);
964 opr_mutex_exit(&vq->mutex);
966 vb = malloc(sizeof(struct volume_init_batch));
968 vb->thread = params->thread;
977 opr_mutex_enter(&vq->mutex);
978 queue_Append(vq, vb);
979 opr_cv_broadcast(&vq->cv);
980 opr_mutex_exit(&vq->mutex);
982 Log("Partition scan thread %d of %d ended\n", params->thread, params->nthreads);
988 * Read next element from the pre-populated partition list.
990 static struct DiskPartition64*
991 VInitNextPartition(struct partition_queue *pq)
993 struct DiskPartition64 *partition;
994 struct diskpartition_queue_t *dp; /* queue element */
996 if (vinit_attach_abort) {
997 Log("Aborting volume preattach thread.\n");
1001 /* get next partition to scan */
1002 opr_mutex_enter(&pq->mutex);
1003 if (queue_IsEmpty(pq)) {
1004 opr_mutex_exit(&pq->mutex);
1007 dp = queue_First(pq, diskpartition_queue_t);
1009 opr_mutex_exit(&pq->mutex);
1012 opr_Assert(dp->diskP);
1014 partition = dp->diskP;
1020 * Find next volume id on the partition.
1023 VInitNextVolumeId(DIR *dirp)
1029 while((d = readdir(dirp))) {
1030 if (vinit_attach_abort) {
1031 Log("Aborting volume preattach thread.\n");
1034 ext = strrchr(d->d_name, '.');
1035 if (d->d_name[0] == 'V' && ext && strcmp(ext, VHDREXT) == 0) {
1036 vid = VolumeNumber(d->d_name);
1040 Log("Warning: bogus volume header file: %s\n", d->d_name);
1047 * Preattach volumes in batches to avoid lock contention.
1050 VInitPreAttachVolumes(int nthreads, struct volume_init_queue *vq)
1052 struct volume_init_batch *vb;
1056 /* dequeue next volume */
1057 opr_mutex_enter(&vq->mutex);
1058 if (queue_IsEmpty(vq)) {
1059 opr_cv_wait(&vq->cv, &vq->mutex);
1061 vb = queue_First(vq, volume_init_batch);
1063 opr_mutex_exit(&vq->mutex);
1067 for (i = 0; i<vb->size; i++) {
1073 dup = VLookupVolume_r(&ec, vp->hashid, NULL);
1075 Log("Error looking up volume, code=%d\n", ec);
1078 Log("Warning: Duplicate volume id %" AFS_VOLID_FMT " detected.\n", afs_printable_VolumeId_lu(vp->hashid));
1081 /* put pre-attached volume onto the hash table
1082 * and bring it up to the pre-attached state */
1083 AddVolumeToHashTable(vp, vp->hashid);
1084 AddVolumeToVByPList_r(vp);
1085 VLRU_Init_Node_r(vp);
1086 VChangeState_r(vp, VOL_STATE_PREATTACHED);
1099 #endif /* AFS_DEMAND_ATTACH_FS */
1101 #if !defined(AFS_DEMAND_ATTACH_FS)
1103 * attach all volumes on a given disk partition
1106 VAttachVolumesByPartition(struct DiskPartition64 *diskP, int * nAttached, int * nUnattached)
1112 Log("Partition %s: attaching volumes\n", diskP->name);
1113 dirp = opendir(VPartitionPath(diskP));
1115 Log("opendir on Partition %s failed!\n", diskP->name);
1119 while ((dp = readdir(dirp))) {
1121 p = strrchr(dp->d_name, '.');
1123 if (vinit_attach_abort) {
1124 Log("Partition %s: abort attach volumes\n", diskP->name);
1128 if (p != NULL && strcmp(p, VHDREXT) == 0) {
1131 vp = VAttachVolumeByName(&error, diskP->name, dp->d_name,
1133 (*(vp ? nAttached : nUnattached))++;
1134 if (error == VOFFLINE)
1135 Log("Volume %d stays offline (/vice/offline/%s exists)\n", VolumeNumber(dp->d_name), dp->d_name);
1136 else if (LogLevel >= 5) {
1137 Log("Partition %s: attached volume %d (%s)\n",
1138 diskP->name, VolumeNumber(dp->d_name),
1147 Log("Partition %s: attached %d volumes; %d volumes not attached\n", diskP->name, *nAttached, *nUnattached);
1152 #endif /* !AFS_DEMAND_ATTACH_FS */
1154 /***************************************************/
1155 /* Shutdown routines */
1156 /***************************************************/
1160 * highly multithreaded volume package shutdown
1162 * with the demand attach fileserver extensions,
1163 * VShutdown has been modified to be multithreaded.
1164 * In order to achieve optimal use of many threads,
1165 * the shutdown code involves one control thread and
1166 * n shutdown worker threads. The control thread
1167 * periodically examines the number of volumes available
1168 * for shutdown on each partition, and produces a worker
1169 * thread allocation schedule. The idea is to eliminate
1170 * redundant scheduling computation on the workers by
1171 * having a single master scheduler.
1173 * The scheduler's objectives are:
1175 * each partition with volumes remaining gets allocated
1176 * at least 1 thread (assuming sufficient threads)
1178 * threads are allocated proportional to the number of
1179 * volumes remaining to be offlined. This ensures that
1180 * the OS I/O scheduler has many requests to elevator
1181 * seek on partitions that will (presumably) take the
1182 * longest amount of time (from now) to finish shutdown
1183 * (3) keep threads busy
1184 * when there are extra threads, they are assigned to
1185 * partitions using a simple round-robin algorithm
1187 * In the future, we may wish to add the ability to adapt
1188 * to the relative performance patterns of each disk
1193 * multi-step shutdown process
1195 * demand attach shutdown is a four-step process. Each
1196 * shutdown "pass" shuts down increasingly more difficult
1197 * volumes. The main purpose is to achieve better cache
1198 * utilization during shutdown.
1201 * shutdown volumes in the unattached, pre-attached
1204 * shutdown attached volumes with cached volume headers
1206 * shutdown all volumes in non-exclusive states
1208 * shutdown all remaining volumes
1211 #ifdef AFS_DEMAND_ATTACH_FS
1217 struct DiskPartition64 * diskP;
1218 struct diskpartition_queue_t * dpq;
1219 vshutdown_thread_t params;
1221 pthread_attr_t attrs;
1223 memset(¶ms, 0, sizeof(vshutdown_thread_t));
1226 Log("VShutdown: aborting attach volumes\n");
1227 vinit_attach_abort = 1;
1228 VOL_CV_WAIT(&vol_init_attach_cond);
1231 for (params.n_parts=0, diskP = DiskPartitionList;
1232 diskP; diskP = diskP->next, params.n_parts++);
1234 Log("VShutdown: shutting down on-line volumes on %d partition%s...\n",
1235 params.n_parts, params.n_parts > 1 ? "s" : "");
1237 vol_shutting_down = 1;
1239 if (vol_attach_threads > 1) {
1240 /* prepare for parallel shutdown */
1241 params.n_threads = vol_attach_threads;
1242 opr_mutex_init(¶ms.lock);
1243 opr_cv_init(¶ms.cv);
1244 opr_cv_init(¶ms.master_cv);
1245 opr_Verify(pthread_attr_init(&attrs) == 0);
1246 opr_Verify(pthread_attr_setdetachstate(&attrs,
1247 PTHREAD_CREATE_DETACHED) == 0);
1248 queue_Init(¶ms);
1250 /* setup the basic partition information structures for
1251 * parallel shutdown */
1252 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1254 struct rx_queue * qp, * nqp;
1258 VVByPListWait_r(diskP);
1259 VVByPListBeginExclusive_r(diskP);
1262 for (queue_Scan(&diskP->vol_list, qp, nqp, rx_queue)) {
1263 vp = (Volume *)((char *)qp - offsetof(Volume, vol_list));
1267 Log("VShutdown: partition %s has %d volumes with attached headers\n",
1268 VPartitionPath(diskP), count);
1271 /* build up the pass 0 shutdown work queue */
1272 dpq = malloc(sizeof(struct diskpartition_queue_t));
1273 opr_Assert(dpq != NULL);
1275 queue_Prepend(¶ms, dpq);
1277 params.part_pass_head[diskP->index] = queue_First(&diskP->vol_list, rx_queue);
1280 Log("VShutdown: beginning parallel fileserver shutdown\n");
1281 Log("VShutdown: using %d threads to offline volumes on %d partition%s\n",
1282 vol_attach_threads, params.n_parts, params.n_parts > 1 ? "s" : "" );
1284 /* do pass 0 shutdown */
1285 opr_mutex_enter(¶ms.lock);
1286 for (i=0; i < params.n_threads; i++) {
1287 opr_Verify(pthread_create(&tid, &attrs, &VShutdownThread,
1291 /* wait for all the pass 0 shutdowns to complete */
1292 while (params.n_threads_complete < params.n_threads) {
1293 CV_WAIT(¶ms.master_cv, ¶ms.lock);
1295 params.n_threads_complete = 0;
1297 opr_cv_broadcast(¶ms.cv);
1298 opr_mutex_exit(¶ms.lock);
1300 Log("VShutdown: pass 0 completed using the 1 thread per partition algorithm\n");
1301 Log("VShutdown: starting passes 1 through 3 using finely-granular mp-fast algorithm\n");
1303 /* run the parallel shutdown scheduler. it will drop the glock internally */
1304 ShutdownController(¶ms);
1306 /* wait for all the workers to finish pass 3 and terminate */
1307 while (params.pass < 4) {
1308 VOL_CV_WAIT(¶ms.cv);
1311 opr_Verify(pthread_attr_destroy(&attrs) == 0);
1312 opr_cv_destroy(¶ms.cv);
1313 opr_cv_destroy(¶ms.master_cv);
1314 opr_mutex_destroy(¶ms.lock);
1316 /* drop the VByPList exclusive reservations */
1317 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1318 VVByPListEndExclusive_r(diskP);
1319 Log("VShutdown: %s stats : (pass[0]=%d, pass[1]=%d, pass[2]=%d, pass[3]=%d)\n",
1320 VPartitionPath(diskP),
1321 params.stats[0][diskP->index],
1322 params.stats[1][diskP->index],
1323 params.stats[2][diskP->index],
1324 params.stats[3][diskP->index]);
1327 Log("VShutdown: shutdown finished using %d threads\n", params.n_threads);
1329 /* if we're only going to run one shutdown thread, don't bother creating
1331 Log("VShutdown: beginning single-threaded fileserver shutdown\n");
1333 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1334 VShutdownByPartition_r(diskP);
1338 Log("VShutdown: complete.\n");
1341 #else /* AFS_DEMAND_ATTACH_FS */
1351 Log("VShutdown: aborting attach volumes\n");
1352 vinit_attach_abort = 1;
1353 #ifdef AFS_PTHREAD_ENV
1354 VOL_CV_WAIT(&vol_init_attach_cond);
1356 LWP_WaitProcess(VInitAttachVolumes);
1357 #endif /* AFS_PTHREAD_ENV */
1360 Log("VShutdown: shutting down on-line volumes...\n");
1361 vol_shutting_down = 1;
1362 for (i = 0; i < VolumeHashTable.Size; i++) {
1363 /* try to hold first volume in the hash table */
1364 for (queue_Scan(&VolumeHashTable.Table[i],vp,np,Volume)) {
1368 Log("VShutdown: Attempting to take volume %" AFS_VOLID_FMT " offline.\n",
1369 afs_printable_VolumeId_lu(vp->hashid));
1371 /* next, take the volume offline (drops reference count) */
1372 VOffline_r(vp, "File server was shut down");
1376 Log("VShutdown: complete.\n");
1378 #endif /* AFS_DEMAND_ATTACH_FS */
1384 opr_Assert(VInit>0);
1391 * stop new activity (e.g. SALVSYNC) from occurring
1393 * Use this to make the volume package less busy; for example, during
1394 * shutdown. This doesn't actually shutdown/detach anything in the
1395 * volume package, but prevents certain processes from ocurring. For
1396 * example, preventing new SALVSYNC communication in DAFS. In theory, we
1397 * could also use this to prevent new volume attachment, or prevent
1398 * other programs from checking out volumes, etc.
1403 #ifdef AFS_DEMAND_ATTACH_FS
1404 /* make sure we don't try to contact the salvageserver, since it may
1405 * not be around anymore */
1406 vol_disallow_salvsync = 1;
1410 #ifdef AFS_DEMAND_ATTACH_FS
1413 * shutdown control thread
1416 ShutdownController(vshutdown_thread_t * params)
1419 struct DiskPartition64 * diskP;
1421 vshutdown_thread_t shadow;
1423 ShutdownCreateSchedule(params);
1425 while ((params->pass < 4) &&
1426 (params->n_threads_complete < params->n_threads)) {
1427 /* recompute schedule once per second */
1429 memcpy(&shadow, params, sizeof(vshutdown_thread_t));
1433 Log("ShutdownController: schedule version=%d, vol_remaining=%d, pass=%d\n",
1434 shadow.schedule_version, shadow.vol_remaining, shadow.pass);
1435 Log("ShutdownController: n_threads_complete=%d, n_parts_done_pass=%d\n",
1436 shadow.n_threads_complete, shadow.n_parts_done_pass);
1437 for (diskP = DiskPartitionList; diskP; diskP=diskP->next) {
1439 Log("ShutdownController: part[%d] : (len=%d, thread_target=%d, done_pass=%d, pass_head=%p)\n",
1441 diskP->vol_list.len,
1442 shadow.part_thread_target[id],
1443 shadow.part_done_pass[id],
1444 shadow.part_pass_head[id]);
1450 ShutdownCreateSchedule(params);
1454 /* create the shutdown thread work schedule.
1455 * this scheduler tries to implement fairness
1456 * by allocating at least 1 thread to each
1457 * partition with volumes to be shutdown,
1458 * and then it attempts to allocate remaining
1459 * threads based upon the amount of work left
1462 ShutdownCreateSchedule(vshutdown_thread_t * params)
1464 struct DiskPartition64 * diskP;
1465 int sum, thr_workload, thr_left;
1466 int part_residue[VOLMAXPARTS+1];
1469 /* compute the total number of outstanding volumes */
1471 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1472 sum += diskP->vol_list.len;
1475 params->schedule_version++;
1476 params->vol_remaining = sum;
1481 /* compute average per-thread workload */
1482 thr_workload = sum / params->n_threads;
1483 if (sum % params->n_threads)
1486 thr_left = params->n_threads;
1487 memset(&part_residue, 0, sizeof(part_residue));
1489 /* for fairness, give every partition with volumes remaining
1490 * at least one thread */
1491 for (diskP = DiskPartitionList; diskP && thr_left; diskP = diskP->next) {
1493 if (diskP->vol_list.len) {
1494 params->part_thread_target[id] = 1;
1497 params->part_thread_target[id] = 0;
1501 if (thr_left && thr_workload) {
1502 /* compute length-weighted workloads */
1505 for (diskP = DiskPartitionList; diskP && thr_left; diskP = diskP->next) {
1507 delta = (diskP->vol_list.len / thr_workload) -
1508 params->part_thread_target[id];
1512 if (delta < thr_left) {
1513 params->part_thread_target[id] += delta;
1516 params->part_thread_target[id] += thr_left;
1524 /* try to assign any leftover threads to partitions that
1525 * had volume lengths closer to needing thread_target+1 */
1526 int max_residue, max_id = 0;
1528 /* compute the residues */
1529 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1531 part_residue[id] = diskP->vol_list.len -
1532 (params->part_thread_target[id] * thr_workload);
1535 /* now try to allocate remaining threads to partitions with the
1536 * highest residues */
1539 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1541 if (part_residue[id] > max_residue) {
1542 max_residue = part_residue[id];
1551 params->part_thread_target[max_id]++;
1553 part_residue[max_id] = 0;
1558 /* punt and give any remaining threads equally to each partition */
1560 if (thr_left >= params->n_parts) {
1561 alloc = thr_left / params->n_parts;
1562 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1564 params->part_thread_target[id] += alloc;
1569 /* finish off the last of the threads */
1570 for (diskP = DiskPartitionList; thr_left && diskP; diskP = diskP->next) {
1572 params->part_thread_target[id]++;
1578 /* worker thread for parallel shutdown */
1580 VShutdownThread(void * args)
1582 vshutdown_thread_t * params;
1583 int found, pass, schedule_version_save, count;
1584 struct DiskPartition64 *diskP;
1585 struct diskpartition_queue_t * dpq;
1588 params = (vshutdown_thread_t *) args;
1590 /* acquire the shutdown pass 0 lock */
1591 opr_mutex_enter(¶ms->lock);
1593 /* if there's still pass 0 work to be done,
1594 * get a work entry, and do a pass 0 shutdown */
1595 if (queue_IsNotEmpty(params)) {
1596 dpq = queue_First(params, diskpartition_queue_t);
1598 opr_mutex_exit(¶ms->lock);
1604 while (ShutdownVolumeWalk_r(diskP, 0, ¶ms->part_pass_head[id]))
1606 params->stats[0][diskP->index] = count;
1607 opr_mutex_enter(¶ms->lock);
1610 params->n_threads_complete++;
1611 if (params->n_threads_complete == params->n_threads) {
1612 /* notify control thread that all workers have completed pass 0 */
1613 opr_cv_signal(¶ms->master_cv);
1615 while (params->pass == 0) {
1616 opr_cv_wait(¶ms->cv, ¶ms->lock);
1620 opr_mutex_exit(¶ms->lock);
1623 pass = params->pass;
1624 opr_Assert(pass > 0);
1626 /* now escalate through the more complicated shutdowns */
1628 schedule_version_save = params->schedule_version;
1630 /* find a disk partition to work on */
1631 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1633 if (params->part_thread_target[id] && !params->part_done_pass[id]) {
1634 params->part_thread_target[id]--;
1641 /* hmm. for some reason the controller thread couldn't find anything for
1642 * us to do. let's see if there's anything we can do */
1643 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1645 if (diskP->vol_list.len && !params->part_done_pass[id]) {
1648 } else if (!params->part_done_pass[id]) {
1649 params->part_done_pass[id] = 1;
1650 params->n_parts_done_pass++;
1652 Log("VShutdown: done shutting down volumes on partition %s.\n",
1653 VPartitionPath(diskP));
1659 /* do work on this partition until either the controller
1660 * creates a new schedule, or we run out of things to do
1661 * on this partition */
1664 while (!params->part_done_pass[id] &&
1665 (schedule_version_save == params->schedule_version)) {
1666 /* ShutdownVolumeWalk_r will drop the glock internally */
1667 if (!ShutdownVolumeWalk_r(diskP, pass, ¶ms->part_pass_head[id])) {
1668 if (!params->part_done_pass[id]) {
1669 params->part_done_pass[id] = 1;
1670 params->n_parts_done_pass++;
1672 Log("VShutdown: done shutting down volumes on partition %s.\n",
1673 VPartitionPath(diskP));
1681 params->stats[pass][id] += count;
1683 /* ok, everyone is done this pass, proceed */
1686 params->n_threads_complete++;
1687 while (params->pass == pass) {
1688 if (params->n_threads_complete == params->n_threads) {
1689 /* we are the last thread to complete, so we will
1690 * reinitialize worker pool state for the next pass */
1691 params->n_threads_complete = 0;
1692 params->n_parts_done_pass = 0;
1694 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1696 params->part_done_pass[id] = 0;
1697 params->part_pass_head[id] = queue_First(&diskP->vol_list, rx_queue);
1700 /* compute a new thread schedule before releasing all the workers */
1701 ShutdownCreateSchedule(params);
1703 /* wake up all the workers */
1704 opr_cv_broadcast(¶ms->cv);
1707 Log("VShutdown: pass %d completed using %d threads on %d partitions\n",
1708 pass, params->n_threads, params->n_parts);
1711 VOL_CV_WAIT(¶ms->cv);
1714 pass = params->pass;
1728 /* shut down all volumes on a given disk partition
1730 * note that this function will not allow mp-fast
1731 * shutdown of a partition */
1733 VShutdownByPartition_r(struct DiskPartition64 * dp)
1739 /* wait for other exclusive ops to finish */
1740 VVByPListWait_r(dp);
1742 /* begin exclusive access */
1743 VVByPListBeginExclusive_r(dp);
1745 /* pick the low-hanging fruit first,
1746 * then do the complicated ones last
1747 * (has the advantage of keeping
1748 * in-use volumes up until the bitter end) */
1749 for (pass = 0, total=0; pass < 4; pass++) {
1750 pass_stats[pass] = ShutdownVByPForPass_r(dp, pass);
1751 total += pass_stats[pass];
1754 /* end exclusive access */
1755 VVByPListEndExclusive_r(dp);
1757 Log("VShutdownByPartition: shut down %d volumes on %s (pass[0]=%d, pass[1]=%d, pass[2]=%d, pass[3]=%d)\n",
1758 total, VPartitionPath(dp), pass_stats[0], pass_stats[1], pass_stats[2], pass_stats[3]);
1763 /* internal shutdown functionality
1765 * for multi-pass shutdown:
1766 * 0 to only "shutdown" {pre,un}attached and error state volumes
1767 * 1 to also shutdown attached volumes w/ volume header loaded
1768 * 2 to also shutdown attached volumes w/o volume header loaded
1769 * 3 to also shutdown exclusive state volumes
1771 * caller MUST hold exclusive access on the hash chain
1772 * because we drop vol_glock_mutex internally
1774 * this function is reentrant for passes 1--3
1775 * (e.g. multiple threads can cooperate to
1776 * shutdown a partition mp-fast)
1778 * pass 0 is not scaleable because the volume state data is
1779 * synchronized by vol_glock mutex, and the locking overhead
1780 * is too high to drop the lock long enough to do linked list
1784 ShutdownVByPForPass_r(struct DiskPartition64 * dp, int pass)
1786 struct rx_queue * q = queue_First(&dp->vol_list, rx_queue);
1789 while (ShutdownVolumeWalk_r(dp, pass, &q))
1795 /* conditionally shutdown one volume on partition dp
1796 * returns 1 if a volume was shutdown in this pass,
1799 ShutdownVolumeWalk_r(struct DiskPartition64 * dp, int pass,
1800 struct rx_queue ** idx)
1802 struct rx_queue *qp, *nqp;
1807 for (queue_ScanFrom(&dp->vol_list, qp, qp, nqp, rx_queue)) {
1808 vp = (Volume *) (((char *)qp) - offsetof(Volume, vol_list));
1812 if ((V_attachState(vp) != VOL_STATE_UNATTACHED) &&
1813 (V_attachState(vp) != VOL_STATE_ERROR) &&
1814 (V_attachState(vp) != VOL_STATE_DELETED) &&
1815 (V_attachState(vp) != VOL_STATE_PREATTACHED)) {
1819 if ((V_attachState(vp) == VOL_STATE_ATTACHED) &&
1820 (vp->header == NULL)) {
1824 if (VIsExclusiveState(V_attachState(vp))) {
1829 DeleteVolumeFromVByPList_r(vp);
1830 VShutdownVolume_r(vp);
1840 * shutdown a specific volume
1842 /* caller MUST NOT hold a heavyweight ref on vp */
1844 VShutdownVolume_r(Volume * vp)
1848 VCreateReservation_r(vp);
1850 if (LogLevel >= 5) {
1851 Log("VShutdownVolume_r: vid=%" AFS_VOLID_FMT ", device=%d, state=%hu\n",
1852 afs_printable_VolumeId_lu(vp->hashid), vp->partition->device,
1856 /* wait for other blocking ops to finish */
1857 VWaitExclusiveState_r(vp);
1859 opr_Assert(VIsValidState(V_attachState(vp)));
1861 switch(V_attachState(vp)) {
1862 case VOL_STATE_SALVAGING:
1863 /* Leave salvaging volumes alone. Any in-progress salvages will
1864 * continue working after viced shuts down. This is intentional.
1867 case VOL_STATE_PREATTACHED:
1868 case VOL_STATE_ERROR:
1869 VChangeState_r(vp, VOL_STATE_UNATTACHED);
1870 case VOL_STATE_UNATTACHED:
1871 case VOL_STATE_DELETED:
1873 case VOL_STATE_GOING_OFFLINE:
1874 case VOL_STATE_SHUTTING_DOWN:
1875 case VOL_STATE_ATTACHED:
1879 Log("VShutdown: Attempting to take volume %" AFS_VOLID_FMT " offline.\n",
1880 afs_printable_VolumeId_lu(vp->hashid));
1882 /* take the volume offline (drops reference count) */
1883 VOffline_r(vp, "File server was shut down");
1890 VCancelReservation_r(vp);
1894 #endif /* AFS_DEMAND_ATTACH_FS */
1897 /***************************************************/
1898 /* Header I/O routines */
1899 /***************************************************/
1902 HeaderName(bit32 magic)
1905 case VOLUMEINFOMAGIC:
1906 return "volume info";
1907 case SMALLINDEXMAGIC:
1908 return "small index";
1909 case LARGEINDEXMAGIC:
1910 return "large index";
1911 case LINKTABLEMAGIC:
1912 return "link table";
1917 /* open a descriptor for the inode (h),
1918 * read in an on-disk structure into buffer (to) of size (size),
1919 * verify versionstamp in structure has magic (magic) and
1920 * optionally verify version (version) if (version) is nonzero
1923 ReadHeader(Error * ec, IHandle_t * h, char *to, int size, bit32 magic,
1926 struct versionStamp *vsn;
1928 afs_sfsize_t nbytes;
1933 Log("ReadHeader: Null inode handle argument for %s header file.\n",
1941 Log("ReadHeader: Failed to open %s header file "
1942 "(volume=%" AFS_VOLID_FMT ", inode=%s); errno=%d\n", HeaderName(magic), afs_printable_VolumeId_lu(h->ih_vid),
1943 PrintInode(stmp, h->ih_ino), errno);
1948 vsn = (struct versionStamp *)to;
1949 nbytes = FDH_PREAD(fdP, to, size, 0);
1951 Log("ReadHeader: Failed to read %s header file "
1952 "(volume=%" AFS_VOLID_FMT ", inode=%s); errno=%d\n", HeaderName(magic), afs_printable_VolumeId_lu(h->ih_vid),
1953 PrintInode(stmp, h->ih_ino), errno);
1955 FDH_REALLYCLOSE(fdP);
1958 if (nbytes != size) {
1959 Log("ReadHeader: Incorrect number of bytes read from %s header file "
1960 "(volume=%" AFS_VOLID_FMT ", inode=%s); expected=%d, read=%d\n",
1961 HeaderName(magic), afs_printable_VolumeId_lu(h->ih_vid),
1962 PrintInode(stmp, h->ih_ino), size, (int)nbytes);
1964 FDH_REALLYCLOSE(fdP);
1967 if (vsn->magic != magic) {
1968 Log("ReadHeader: Incorrect magic for %s header file "
1969 "(volume=%" AFS_VOLID_FMT ", inode=%s); expected=0x%x, read=0x%x\n",
1970 HeaderName(magic), afs_printable_VolumeId_lu(h->ih_vid),
1971 PrintInode(stmp, h->ih_ino), magic, vsn->magic);
1973 FDH_REALLYCLOSE(fdP);
1979 /* Check is conditional, in case caller wants to inspect version himself */
1980 if (version && vsn->version != version) {
1981 Log("ReadHeader: Incorrect version for %s header file "
1982 "(volume=%" AFS_VOLID_FMT ", inode=%s); expected=%x, read=%x\n",
1983 HeaderName(magic), afs_printable_VolumeId_lu(h->ih_vid), PrintInode(stmp, h->ih_ino),
1984 version, vsn->version);
1990 WriteVolumeHeader_r(Error * ec, Volume * vp)
1992 IHandle_t *h = V_diskDataHandle(vp);
2002 if (FDH_PWRITE(fdP, (char *)&V_disk(vp), sizeof(V_disk(vp)), 0)
2003 != sizeof(V_disk(vp))) {
2005 FDH_REALLYCLOSE(fdP);
2011 /* VolumeHeaderToDisk
2012 * Allows for storing 64 bit inode numbers in on-disk volume header
2015 /* convert in-memory representation of a volume header to the
2016 * on-disk representation of a volume header */
2018 VolumeHeaderToDisk(VolumeDiskHeader_t * dh, VolumeHeader_t * h)
2021 memset(dh, 0, sizeof(VolumeDiskHeader_t));
2022 dh->stamp = h->stamp;
2024 dh->parent = h->parent;
2026 #ifdef AFS_64BIT_IOPS_ENV
2027 dh->volumeInfo_lo = (afs_int32) h->volumeInfo & 0xffffffff;
2028 dh->volumeInfo_hi = (afs_int32) (h->volumeInfo >> 32) & 0xffffffff;
2029 dh->smallVnodeIndex_lo = (afs_int32) h->smallVnodeIndex & 0xffffffff;
2030 dh->smallVnodeIndex_hi =
2031 (afs_int32) (h->smallVnodeIndex >> 32) & 0xffffffff;
2032 dh->largeVnodeIndex_lo = (afs_int32) h->largeVnodeIndex & 0xffffffff;
2033 dh->largeVnodeIndex_hi =
2034 (afs_int32) (h->largeVnodeIndex >> 32) & 0xffffffff;
2035 dh->linkTable_lo = (afs_int32) h->linkTable & 0xffffffff;
2036 dh->linkTable_hi = (afs_int32) (h->linkTable >> 32) & 0xffffffff;
2038 dh->volumeInfo_lo = h->volumeInfo;
2039 dh->smallVnodeIndex_lo = h->smallVnodeIndex;
2040 dh->largeVnodeIndex_lo = h->largeVnodeIndex;
2041 dh->linkTable_lo = h->linkTable;
2045 /* DiskToVolumeHeader
2046 * Converts an on-disk representation of a volume header to
2047 * the in-memory representation of a volume header.
2049 * Makes the assumption that AFS has *always*
2050 * zero'd the volume header file so that high parts of inode
2051 * numbers are 0 in older (SGI EFS) volume header files.
2054 DiskToVolumeHeader(VolumeHeader_t * h, VolumeDiskHeader_t * dh)
2056 memset(h, 0, sizeof(VolumeHeader_t));
2057 h->stamp = dh->stamp;
2059 h->parent = dh->parent;
2061 #ifdef AFS_64BIT_IOPS_ENV
2063 (Inode) dh->volumeInfo_lo | ((Inode) dh->volumeInfo_hi << 32);
2065 h->smallVnodeIndex =
2066 (Inode) dh->smallVnodeIndex_lo | ((Inode) dh->
2067 smallVnodeIndex_hi << 32);
2069 h->largeVnodeIndex =
2070 (Inode) dh->largeVnodeIndex_lo | ((Inode) dh->
2071 largeVnodeIndex_hi << 32);
2073 (Inode) dh->linkTable_lo | ((Inode) dh->linkTable_hi << 32);
2075 h->volumeInfo = dh->volumeInfo_lo;
2076 h->smallVnodeIndex = dh->smallVnodeIndex_lo;
2077 h->largeVnodeIndex = dh->largeVnodeIndex_lo;
2078 h->linkTable = dh->linkTable_lo;
2083 /***************************************************/
2084 /* Volume Attachment routines */
2085 /***************************************************/
2087 #ifdef AFS_DEMAND_ATTACH_FS
2089 * pre-attach a volume given its path.
2091 * @param[out] ec outbound error code
2092 * @param[in] partition partition path string
2093 * @param[in] name volume id string
2095 * @return volume object pointer
2097 * @note A pre-attached volume will only have its partition
2098 * and hashid fields initialized. At first call to
2099 * VGetVolume, the volume will be fully attached.
2103 VPreAttachVolumeByName(Error * ec, char *partition, char *name)
2107 vp = VPreAttachVolumeByName_r(ec, partition, name);
2113 * pre-attach a volume given its path.
2115 * @param[out] ec outbound error code
2116 * @param[in] partition path to vice partition
2117 * @param[in] name volume id string
2119 * @return volume object pointer
2121 * @pre VOL_LOCK held
2123 * @internal volume package internal use only.
2126 VPreAttachVolumeByName_r(Error * ec, char *partition, char *name)
2128 return VPreAttachVolumeById_r(ec,
2130 VolumeNumber(name));
2134 * pre-attach a volume given its path and numeric volume id.
2136 * @param[out] ec error code return
2137 * @param[in] partition path to vice partition
2138 * @param[in] volumeId numeric volume id
2140 * @return volume object pointer
2142 * @pre VOL_LOCK held
2144 * @internal volume package internal use only.
2147 VPreAttachVolumeById_r(Error * ec,
2152 struct DiskPartition64 *partp;
2156 opr_Assert(programType == fileServer);
2158 if (!(partp = VGetPartition_r(partition, 0))) {
2160 Log("VPreAttachVolumeById_r: Error getting partition (%s)\n", partition);
2164 /* ensure that any vp we pass to VPreAttachVolumeByVp_r
2165 * is NOT in exclusive state.
2168 vp = VLookupVolume_r(ec, volumeId, NULL);
2174 if (vp && VIsExclusiveState(V_attachState(vp))) {
2175 VCreateReservation_r(vp);
2176 VWaitExclusiveState_r(vp);
2177 VCancelReservation_r(vp);
2179 goto retry; /* look up volume again */
2182 /* vp == NULL or vp not exclusive both OK */
2184 return VPreAttachVolumeByVp_r(ec, partp, vp, volumeId);
2188 * preattach a volume.
2190 * @param[out] ec outbound error code
2191 * @param[in] partp pointer to partition object
2192 * @param[in] vp pointer to volume object
2193 * @param[in] vid volume id
2195 * @return volume object pointer
2197 * @pre VOL_LOCK is held.
2199 * @pre vp (if specified) must not be in exclusive state.
2201 * @warning Returned volume object pointer does not have to
2202 * equal the pointer passed in as argument vp. There
2203 * are potential race conditions which can result in
2204 * the pointers having different values. It is up to
2205 * the caller to make sure that references are handled
2206 * properly in this case.
2208 * @note If there is already a volume object registered with
2209 * the same volume id, its pointer MUST be passed as
2210 * argument vp. Failure to do so will result in a silent
2211 * failure to preattach.
2213 * @internal volume package internal use only.
2216 VPreAttachVolumeByVp_r(Error * ec,
2217 struct DiskPartition64 * partp,
2225 /* don't proceed unless it's safe */
2227 opr_Assert(!VIsExclusiveState(V_attachState(vp)));
2230 /* check to see if pre-attach already happened */
2232 (V_attachState(vp) != VOL_STATE_UNATTACHED) &&
2233 (V_attachState(vp) != VOL_STATE_DELETED) &&
2234 (V_attachState(vp) != VOL_STATE_PREATTACHED) &&
2235 !VIsErrorState(V_attachState(vp))) {
2237 * pre-attach is a no-op in all but the following cases:
2239 * - volume is unattached
2240 * - volume is in an error state
2241 * - volume is pre-attached
2243 Log("VPreattachVolumeByVp_r: volume %" AFS_VOLID_FMT " not in quiescent state (state %u flags 0x%x)\n",
2244 afs_printable_VolumeId_lu(vid), V_attachState(vp),
2248 /* we're re-attaching a volume; clear out some old state */
2249 memset(&vp->salvage, 0, sizeof(struct VolumeOnlineSalvage));
2251 if (V_partition(vp) != partp) {
2252 /* XXX potential race */
2253 DeleteVolumeFromVByPList_r(vp);
2256 /* if we need to allocate a new Volume struct,
2257 * go ahead and drop the vol glock, otherwise
2258 * do the basic setup synchronised, as it's
2259 * probably not worth dropping the lock */
2262 /* allocate the volume structure */
2263 vp = nvp = calloc(1, sizeof(Volume));
2264 opr_Assert(vp != NULL);
2265 queue_Init(&vp->vnode_list);
2266 queue_Init(&vp->rx_call_list);
2267 opr_cv_init(&V_attachCV(vp));
2270 /* link the volume with its associated vice partition */
2271 vp->device = partp->device;
2272 vp->partition = partp;
2275 vp->specialStatus = 0;
2277 /* if we dropped the lock, reacquire the lock,
2278 * check for pre-attach races, and then add
2279 * the volume to the hash table */
2282 nvp = VLookupVolume_r(ec, vid, NULL);
2287 } else if (nvp) { /* race detected */
2292 /* hack to make up for VChangeState_r() decrementing
2293 * the old state counter */
2294 VStats.state_levels[0]++;
2298 /* put pre-attached volume onto the hash table
2299 * and bring it up to the pre-attached state */
2300 AddVolumeToHashTable(vp, vp->hashid);
2301 AddVolumeToVByPList_r(vp);
2302 VLRU_Init_Node_r(vp);
2303 VChangeState_r(vp, VOL_STATE_PREATTACHED);
2306 Log("VPreAttachVolumeByVp_r: volume %" AFS_VOLID_FMT " pre-attached\n", afs_printable_VolumeId_lu(vp->hashid));
2314 #endif /* AFS_DEMAND_ATTACH_FS */
2316 /* Attach an existing volume, given its pathname, and return a
2317 pointer to the volume header information. The volume also
2318 normally goes online at this time. An offline volume
2319 must be reattached to make it go online */
2321 VAttachVolumeByName(Error * ec, char *partition, char *name, int mode)
2325 retVal = VAttachVolumeByName_r(ec, partition, name, mode);
2331 VAttachVolumeByName_r(Error * ec, char *partition, char *name, int mode)
2334 struct DiskPartition64 *partp;
2339 #ifdef AFS_DEMAND_ATTACH_FS
2340 VolumeStats stats_save;
2342 #endif /* AFS_DEMAND_ATTACH_FS */
2346 volumeId = VolumeNumber(name);
2348 if (!(partp = VGetPartition_r(partition, 0))) {
2350 Log("VAttachVolume: Error getting partition (%s)\n", partition);
2354 if (VRequiresPartLock()) {
2355 opr_Assert(VInit == 3);
2356 VLockPartition_r(partition);
2357 } else if (programType == fileServer) {
2358 #ifdef AFS_DEMAND_ATTACH_FS
2359 /* lookup the volume in the hash table */
2360 vp = VLookupVolume_r(ec, volumeId, NULL);
2366 /* save any counters that are supposed to
2367 * be monotonically increasing over the
2368 * lifetime of the fileserver */
2369 memcpy(&stats_save, &vp->stats, sizeof(VolumeStats));
2371 memset(&stats_save, 0, sizeof(VolumeStats));
2374 /* if there's something in the hash table, and it's not
2375 * in the pre-attach state, then we may need to detach
2376 * it before proceeding */
2377 if (vp && (V_attachState(vp) != VOL_STATE_PREATTACHED)) {
2378 VCreateReservation_r(vp);
2379 VWaitExclusiveState_r(vp);
2381 /* at this point state must be one of:
2391 if (vp->specialStatus == VBUSY)
2394 /* if it's already attached, see if we can return it */
2395 if (V_attachState(vp) == VOL_STATE_ATTACHED) {
2396 VGetVolumeByVp_r(ec, vp);
2397 if (V_inUse(vp) == fileServer) {
2398 VCancelReservation_r(vp);
2402 /* otherwise, we need to detach, and attempt to re-attach */
2403 VDetachVolume_r(ec, vp);
2405 Log("VAttachVolume: Error detaching old volume instance (%s)\n", name);
2408 /* if it isn't fully attached, delete from the hash tables,
2409 and let the refcounter handle the rest */
2410 DeleteVolumeFromHashTable(vp);
2411 DeleteVolumeFromVByPList_r(vp);
2414 VCancelReservation_r(vp);
2418 /* pre-attach volume if it hasn't been done yet */
2420 (V_attachState(vp) == VOL_STATE_UNATTACHED) ||
2421 (V_attachState(vp) == VOL_STATE_DELETED) ||
2422 (V_attachState(vp) == VOL_STATE_ERROR)) {
2424 vp = VPreAttachVolumeByVp_r(ec, partp, vp, volumeId);
2430 opr_Assert(vp != NULL);
2432 /* handle pre-attach races
2434 * multiple threads can race to pre-attach a volume,
2435 * but we can't let them race beyond that
2437 * our solution is to let the first thread to bring
2438 * the volume into an exclusive state win; the other
2439 * threads just wait until it finishes bringing the
2440 * volume online, and then they do a vgetvolumebyvp
2442 if (svp && (svp != vp)) {
2443 /* wait for other exclusive ops to finish */
2444 VCreateReservation_r(vp);
2445 VWaitExclusiveState_r(vp);
2447 /* get a heavyweight ref, kill the lightweight ref, and return */
2448 VGetVolumeByVp_r(ec, vp);
2449 VCancelReservation_r(vp);
2453 /* at this point, we are chosen as the thread to do
2454 * demand attachment for this volume. all other threads
2455 * doing a getvolume on vp->hashid will block until we finish */
2457 /* make sure any old header cache entries are invalidated
2458 * before proceeding */
2459 FreeVolumeHeader(vp);
2461 VChangeState_r(vp, VOL_STATE_ATTACHING);
2463 /* restore any saved counters */
2464 memcpy(&vp->stats, &stats_save, sizeof(VolumeStats));
2465 #else /* AFS_DEMAND_ATTACH_FS */
2466 vp = VGetVolume_r(ec, volumeId);
2468 if (V_inUse(vp) == fileServer)
2470 if (vp->specialStatus == VBUSY)
2472 VDetachVolume_r(ec, vp);
2474 Log("VAttachVolume: Error detaching volume (%s)\n", name);
2478 #endif /* AFS_DEMAND_ATTACH_FS */
2482 strcpy(path, VPartitionPath(partp));
2486 strcat(path, OS_DIRSEP);
2490 vp = (Volume *) calloc(1, sizeof(Volume));
2491 opr_Assert(vp != NULL);
2492 vp->hashid = volumeId;
2493 vp->device = partp->device;
2494 vp->partition = partp;
2495 queue_Init(&vp->vnode_list);
2496 queue_Init(&vp->rx_call_list);
2497 #ifdef AFS_DEMAND_ATTACH_FS
2498 opr_cv_init(&V_attachCV(vp));
2499 #endif /* AFS_DEMAND_ATTACH_FS */
2502 /* attach2 is entered without any locks, and returns
2503 * with vol_glock_mutex held */
2504 vp = attach2(ec, volumeId, path, partp, vp, isbusy, mode, &checkedOut);
2506 if (VCanUseFSSYNC() && vp) {
2507 #ifdef AFS_DEMAND_ATTACH_FS
2508 if ((mode == V_VOLUPD) || (VolumeWriteable(vp) && (mode == V_CLONE))) {
2509 /* mark volume header as in use so that volser crashes lead to a
2510 * salvage attempt */
2511 VUpdateVolume_r(ec, vp, 0);
2513 /* for dafs, we should tell the fileserver, except for V_PEEK
2514 * where we know it is not necessary */
2515 if (mode == V_PEEK) {
2516 vp->needsPutBack = 0;
2518 vp->needsPutBack = VOL_PUTBACK;
2520 #else /* !AFS_DEMAND_ATTACH_FS */
2521 /* duplicate computation in fssync.c about whether the server
2522 * takes the volume offline or not. If the volume isn't
2523 * offline, we must not return it when we detach the volume,
2524 * or the server will abort */
2525 if (mode == V_READONLY || mode == V_PEEK
2526 || (!VolumeWriteable(vp) && (mode == V_CLONE || mode == V_DUMP)))
2527 vp->needsPutBack = 0;
2529 vp->needsPutBack = VOL_PUTBACK;
2530 #endif /* !AFS_DEMAND_ATTACH_FS */
2532 #ifdef FSSYNC_BUILD_CLIENT
2533 /* Only give back the vol to the fileserver if we checked it out; attach2
2534 * will set checkedOut only if we successfully checked it out from the
2536 if (VCanUseFSSYNC() && vp == NULL && checkedOut) {
2538 #ifdef AFS_DEMAND_ATTACH_FS
2539 /* If we couldn't attach but we scheduled a salvage, we already
2540 * notified the fileserver; don't online it now */
2541 if (*ec != VSALVAGING)
2542 #endif /* AFS_DEMAND_ATTACH_FS */
2543 FSYNC_VolOp(volumeId, partition, FSYNC_VOL_ON, 0, NULL);
2546 if (programType == fileServer && vp) {
2547 #ifdef AFS_DEMAND_ATTACH_FS
2549 * we can get here in cases where we don't "own"
2550 * the volume (e.g. volume owned by a utility).
2551 * short circuit around potential disk header races.
2553 if (V_attachState(vp) != VOL_STATE_ATTACHED) {
2557 VUpdateVolume_r(ec, vp, 0);
2559 Log("VAttachVolume: Error updating volume\n");
2564 if (VolumeWriteable(vp) && V_dontSalvage(vp) == 0) {
2565 #ifndef AFS_DEMAND_ATTACH_FS
2566 /* This is a hack: by temporarily setting the incore
2567 * dontSalvage flag ON, the volume will be put back on the
2568 * Update list (with dontSalvage OFF again). It will then
2569 * come back in N minutes with DONT_SALVAGE eventually
2570 * set. This is the way that volumes that have never had
2571 * it set get it set; or that volumes that have been
2572 * offline without DONT SALVAGE having been set also
2573 * eventually get it set */
2574 V_dontSalvage(vp) = DONT_SALVAGE;
2575 #endif /* !AFS_DEMAND_ATTACH_FS */
2576 VAddToVolumeUpdateList_r(ec, vp);
2578 Log("VAttachVolume: Error adding volume to update list\n");
2585 Log("VOnline: volume %" AFS_VOLID_FMT " (%s) attached and online\n", afs_printable_VolumeId_lu(V_id(vp)),
2590 if (VRequiresPartLock()) {
2591 VUnlockPartition_r(partition);
2594 #ifdef AFS_DEMAND_ATTACH_FS
2595 /* attach failed; make sure we're in error state */
2596 if (vp && !VIsErrorState(V_attachState(vp))) {
2597 VChangeState_r(vp, VOL_STATE_ERROR);
2599 #endif /* AFS_DEMAND_ATTACH_FS */
2606 #ifdef AFS_DEMAND_ATTACH_FS
2607 /* VAttachVolumeByVp_r
2609 * finish attaching a volume that is
2610 * in a less than fully attached state
2612 /* caller MUST hold a ref count on vp */
2614 VAttachVolumeByVp_r(Error * ec, Volume * vp, int mode)
2616 char name[VMAXPATHLEN];
2618 struct DiskPartition64 *partp;
2622 Volume * nvp = NULL;
2623 VolumeStats stats_save;
2627 /* volume utility should never call AttachByVp */
2628 opr_Assert(programType == fileServer);
2630 volumeId = vp->hashid;
2631 partp = vp->partition;
2632 VolumeExternalName_r(volumeId, name, sizeof(name));
2635 /* if another thread is performing a blocking op, wait */
2636 VWaitExclusiveState_r(vp);
2638 memcpy(&stats_save, &vp->stats, sizeof(VolumeStats));
2640 /* if it's already attached, see if we can return it */
2641 if (V_attachState(vp) == VOL_STATE_ATTACHED) {
2642 VGetVolumeByVp_r(ec, vp);
2643 if (V_inUse(vp) == fileServer) {
2646 if (vp->specialStatus == VBUSY)
2648 VDetachVolume_r(ec, vp);
2650 Log("VAttachVolume: Error detaching volume (%s)\n", name);
2656 /* pre-attach volume if it hasn't been done yet */
2658 (V_attachState(vp) == VOL_STATE_UNATTACHED) ||
2659 (V_attachState(vp) == VOL_STATE_DELETED) ||
2660 (V_attachState(vp) == VOL_STATE_ERROR)) {
2661 nvp = VPreAttachVolumeByVp_r(ec, partp, vp, volumeId);
2667 VCreateReservation_r(nvp);
2672 opr_Assert(vp != NULL);
2673 VChangeState_r(vp, VOL_STATE_ATTACHING);
2675 /* restore monotonically increasing stats */
2676 memcpy(&vp->stats, &stats_save, sizeof(VolumeStats));
2680 /* compute path to disk header */
2681 strcpy(path, VPartitionPath(partp));
2685 strcat(path, OS_DIRSEP);
2690 * NOTE: attach2 is entered without any locks, and returns
2691 * with vol_glock_mutex held */
2692 vp = attach2(ec, volumeId, path, partp, vp, isbusy, mode, &checkedOut);
2695 * the event that an error was encountered, or
2696 * the volume was not brought to an attached state
2697 * for any reason, skip to the end. We cannot
2698 * safely call VUpdateVolume unless we "own" it.
2702 (V_attachState(vp) != VOL_STATE_ATTACHED)) {
2706 VUpdateVolume_r(ec, vp, 0);
2708 Log("VAttachVolume: Error updating volume %" AFS_VOLID_FMT "\n",
2709 afs_printable_VolumeId_lu(vp->hashid));
2713 if (VolumeWriteable(vp) && V_dontSalvage(vp) == 0) {
2714 #ifndef AFS_DEMAND_ATTACH_FS
2715 /* This is a hack: by temporarily setting the incore
2716 * dontSalvage flag ON, the volume will be put back on the
2717 * Update list (with dontSalvage OFF again). It will then
2718 * come back in N minutes with DONT_SALVAGE eventually
2719 * set. This is the way that volumes that have never had
2720 * it set get it set; or that volumes that have been
2721 * offline without DONT SALVAGE having been set also
2722 * eventually get it set */
2723 V_dontSalvage(vp) = DONT_SALVAGE;
2724 #endif /* !AFS_DEMAND_ATTACH_FS */
2725 VAddToVolumeUpdateList_r(ec, vp);
2727 Log("VAttachVolume: Error adding volume %" AFS_VOLID_FMT " to update list\n",
2728 afs_printable_VolumeId_lu(vp->hashid));
2735 Log("VOnline: volume %" AFS_VOLID_FMT " (%s) attached and online\n",
2736 afs_printable_VolumeId_lu(V_id(vp)), V_name(vp));
2739 VCancelReservation_r(nvp);
2742 if (*ec && (*ec != VOFFLINE) && (*ec != VSALVAGE)) {
2743 if (vp && !VIsErrorState(V_attachState(vp))) {
2744 VChangeState_r(vp, VOL_STATE_ERROR);
2753 * lock a volume on disk (non-blocking).
2755 * @param[in] vp The volume to lock
2756 * @param[in] locktype READ_LOCK or WRITE_LOCK
2758 * @return operation status
2759 * @retval 0 success, lock was obtained
2760 * @retval EBUSY a conflicting lock was held by another process
2761 * @retval EIO error acquiring lock
2763 * @pre If we're in the fileserver, vp is in an exclusive state
2765 * @pre vp is not already locked
2768 VLockVolumeNB(Volume *vp, int locktype)
2772 opr_Assert(programType != fileServer
2773 || VIsExclusiveState(V_attachState(vp)));
2774 opr_Assert(!(V_attachFlags(vp) & VOL_LOCKED));
2776 code = VLockVolumeByIdNB(vp->hashid, vp->partition, locktype);
2778 V_attachFlags(vp) |= VOL_LOCKED;
2785 * unlock a volume on disk that was locked with VLockVolumeNB.
2787 * @param[in] vp volume to unlock
2789 * @pre If we're in the fileserver, vp is in an exclusive state
2791 * @pre vp has already been locked
2794 VUnlockVolume(Volume *vp)
2796 opr_Assert(programType != fileServer
2797 || VIsExclusiveState(V_attachState(vp)));
2798 opr_Assert((V_attachFlags(vp) & VOL_LOCKED));
2800 VUnlockVolumeById(vp->hashid, vp->partition);
2802 V_attachFlags(vp) &= ~VOL_LOCKED;
2804 #endif /* AFS_DEMAND_ATTACH_FS */
2807 * read in a vol header, possibly lock the vol header, and possibly check out
2808 * the vol header from the fileserver, as part of volume attachment.
2810 * @param[out] ec error code
2811 * @param[in] vp volume pointer object
2812 * @param[in] partp disk partition object of the attaching partition
2813 * @param[in] mode attachment mode such as V_VOLUPD, V_DUMP, etc (see
2815 * @param[in] peek 1 to just try to read in the volume header and make sure
2816 * we don't try to lock the vol, or check it out from
2817 * FSSYNC or anything like that; 0 otherwise, for 'normal'
2819 * @param[out] acheckedOut If we successfully checked-out the volume from
2820 * the fileserver (if we needed to), this is set
2821 * to 1, otherwise it is untouched.
2823 * @note As part of DAFS volume attachment, the volume header may be either
2824 * read- or write-locked to ensure mutual exclusion of certain volume
2825 * operations. In some cases in order to determine whether we need to
2826 * read- or write-lock the header, we need to read in the header to see
2827 * if the volume is RW or not. So, if we read in the header under a
2828 * read-lock and determine that we actually need a write-lock on the
2829 * volume header, this function will drop the read lock, acquire a write
2830 * lock, and read the header in again.
2833 attach_volume_header(Error *ec, Volume *vp, struct DiskPartition64 *partp,
2834 int mode, int peek, int *acheckedOut)
2836 struct VolumeDiskHeader diskHeader;
2837 struct VolumeHeader header;
2840 int lock_tries = 0, checkout_tries = 0;
2842 VolumeId volid = vp->hashid;
2843 #ifdef FSSYNC_BUILD_CLIENT
2844 int checkout, done_checkout = 0;
2845 #endif /* FSSYNC_BUILD_CLIENT */
2846 #ifdef AFS_DEMAND_ATTACH_FS
2847 int locktype = 0, use_locktype = -1;
2848 #endif /* AFS_DEMAND_ATTACH_FS */
2854 if (lock_tries > VOL_MAX_CHECKOUT_RETRIES) {
2855 Log("VAttachVolume: retried too many times trying to lock header for "
2856 "vol %lu part %s; giving up\n", afs_printable_uint32_lu(volid),
2857 VPartitionPath(partp));
2861 if (checkout_tries > VOL_MAX_CHECKOUT_RETRIES) {
2862 Log("VAttachVolume: retried too many times trying to checkout "
2863 "vol %lu part %s; giving up\n", afs_printable_uint32_lu(volid),
2864 VPartitionPath(partp));
2869 if (VReadVolumeDiskHeader(volid, partp, NULL)) {
2870 /* short-circuit the 'volume does not exist' case */
2875 #ifdef FSSYNC_BUILD_CLIENT
2876 checkout = !done_checkout;
2878 if (!peek && checkout && VMustCheckoutVolume(mode)) {
2880 memset(&res, 0, sizeof(res));
2882 if (FSYNC_VolOp(volid, partp->name, FSYNC_VOL_NEEDVOLUME, mode, &res)
2885 if (res.hdr.reason == FSYNC_SALVAGE) {
2886 Log("VAttachVolume: file server says volume %lu is salvaging\n",
2887 afs_printable_uint32_lu(volid));
2890 Log("VAttachVolume: attach of volume %lu apparently denied by file server\n",
2891 afs_printable_uint32_lu(volid));
2892 *ec = VNOVOL; /* XXXX */
2900 #ifdef AFS_DEMAND_ATTACH_FS
2901 if (use_locktype < 0) {
2902 /* don't know whether vol is RO or RW; assume it's RO and we can retry
2903 * if it turns out to be RW */
2904 locktype = VVolLockType(mode, 0);
2907 /* a previous try says we should use use_locktype to lock the volume,
2909 locktype = use_locktype;
2912 if (!peek && locktype) {
2913 code = VLockVolumeNB(vp, locktype);
2915 if (code == EBUSY) {
2916 Log("VAttachVolume: another program has vol %lu locked\n",
2917 afs_printable_uint32_lu(volid));
2919 Log("VAttachVolume: error %d trying to lock vol %lu\n",
2920 code, afs_printable_uint32_lu(volid));
2927 #endif /* AFS_DEMAND_ATTACH_FS */
2929 code = VReadVolumeDiskHeader(volid, partp, &diskHeader);
2939 DiskToVolumeHeader(&header, &diskHeader);
2941 IH_INIT(vp->vnodeIndex[vLarge].handle, partp->device, header.parent,
2942 header.largeVnodeIndex);
2943 IH_INIT(vp->vnodeIndex[vSmall].handle, partp->device, header.parent,
2944 header.smallVnodeIndex);
2945 IH_INIT(vp->diskDataHandle, partp->device, header.parent,
2947 IH_INIT(vp->linkHandle, partp->device, header.parent, header.linkTable);
2950 /* only need to do this once */
2952 GetVolumeHeader(vp);
2956 #if defined(AFS_DEMAND_ATTACH_FS) && defined(FSSYNC_BUILD_CLIENT)
2957 /* demand attach changes the V_PEEK mechanism
2959 * we can now suck the current disk data structure over
2960 * the fssync interface without going to disk
2962 * (technically, we don't need to restrict this feature
2963 * to demand attach fileservers. However, I'm trying
2964 * to limit the number of common code changes)
2966 if (VCanUseFSSYNC() && (mode == V_PEEK || peek)) {
2968 res.payload.len = sizeof(VolumeDiskData);
2969 res.payload.buf = &(V_disk(vp));
2971 if (FSYNC_VolOp(vp->hashid,
2973 FSYNC_VOL_QUERY_HDR,
2976 goto disk_header_loaded;
2979 #endif /* AFS_DEMAND_ATTACH_FS && FSSYNC_BUILD_CLIENT */
2980 (void)ReadHeader(ec, V_diskDataHandle(vp), (char *)&V_disk(vp),
2981 sizeof(V_disk(vp)), VOLUMEINFOMAGIC, VOLUMEINFOVERSION);
2983 #ifdef AFS_DEMAND_ATTACH_FS
2986 IncUInt64(&VStats.hdr_loads);
2987 IncUInt64(&vp->stats.hdr_loads);
2989 #endif /* AFS_DEMAND_ATTACH_FS */
2992 Log("VAttachVolume: Error reading diskDataHandle header for vol %lu; "
2993 "error=%u\n", afs_printable_uint32_lu(volid), *ec);
2997 #ifdef AFS_DEMAND_ATTACH_FS
2998 # ifdef FSSYNC_BUILD_CLIENT
3000 # endif /* FSSYNC_BUILD_CLIENT */
3002 /* if the lock type we actually used to lock the volume is different than
3003 * the lock type we should have used, retry with the lock type we should
3005 use_locktype = VVolLockType(mode, VolumeWriteable(vp));
3006 if (locktype != use_locktype) {
3010 #endif /* AFS_DEMAND_ATTACH_FS */
3015 #if defined(AFS_DEMAND_ATTACH_FS) && defined(FSSYNC_BUILD_CLIENT)
3016 if (!peek && *ec == 0 && retry == 0 && VMustCheckoutVolume(mode)) {
3018 code = FSYNC_VerifyCheckout(volid, partp->name, FSYNC_VOL_NEEDVOLUME, mode);
3020 if (code == SYNC_DENIED) {
3021 /* must retry checkout; fileserver no longer thinks we have
3027 } else if (code != SYNC_OK) {
3031 #endif /* AFS_DEMAND_ATTACH_FS && FSSYNC_BUILD_CLIENT */
3034 /* either we are going to be called again for a second pass, or we
3035 * encountered an error; clean up in either case */
3037 #ifdef AFS_DEMAND_ATTACH_FS
3038 if ((V_attachFlags(vp) & VOL_LOCKED)) {
3041 #endif /* AFS_DEMAND_ATTACH_FS */
3042 if (vp->linkHandle) {
3043 IH_RELEASE(vp->vnodeIndex[vLarge].handle);
3044 IH_RELEASE(vp->vnodeIndex[vSmall].handle);
3045 IH_RELEASE(vp->diskDataHandle);
3046 IH_RELEASE(vp->linkHandle);
3052 FreeVolumeHeader(vp);
3062 #ifdef AFS_DEMAND_ATTACH_FS
3064 attach_check_vop(Error *ec, VolumeId volid, struct DiskPartition64 *partp,
3065 Volume *vp, int *acheckedOut)
3069 if (vp->pending_vol_op) {
3073 if (vp->pending_vol_op->vol_op_state == FSSYNC_VolOpRunningUnknown) {
3075 code = VVolOpLeaveOnlineNoHeader_r(vp, vp->pending_vol_op);
3077 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOnline;
3078 } else if (code == 0) {
3079 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOffline;
3082 /* we need the vol header to determine if the volume can be
3083 * left online for the vop, so... get the header */
3087 /* attach header with peek=1 to avoid checking out the volume
3088 * or locking it; we just want the header info, we're not
3089 * messing with the volume itself at all */
3090 attach_volume_header(ec, vp, partp, V_PEEK, 1, acheckedOut);
3097 if (VVolOpLeaveOnline_r(vp, vp->pending_vol_op)) {
3098 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOnline;
3100 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOffline;
3103 /* make sure we grab a new vol header and re-open stuff on
3104 * actual attachment; we can't keep the data we grabbed, since
3105 * it was not done under a lock and thus not safe */
3106 FreeVolumeHeader(vp);
3107 VReleaseVolumeHandles_r(vp);
3110 /* see if the pending volume op requires exclusive access */
3111 switch (vp->pending_vol_op->vol_op_state) {
3112 case FSSYNC_VolOpPending:
3113 /* this should never happen */
3114 opr_Assert(vp->pending_vol_op->vol_op_state
3115 != FSSYNC_VolOpPending);
3118 case FSSYNC_VolOpRunningUnknown:
3119 /* this should never happen; we resolved 'unknown' above */
3120 opr_Assert(vp->pending_vol_op->vol_op_state
3121 != FSSYNC_VolOpRunningUnknown);
3124 case FSSYNC_VolOpRunningOffline:
3125 /* mark the volume down */
3127 VChangeState_r(vp, VOL_STATE_UNATTACHED);
3129 /* do not set V_offlineMessage here; we don't have ownership of
3130 * the volume (and probably do not have the header loaded), so we
3131 * can't alter the disk header */
3133 /* check to see if we should set the specialStatus flag */
3134 if (VVolOpSetVBusy_r(vp, vp->pending_vol_op)) {
3135 /* don't overwrite specialStatus if it was already set to
3136 * something else (e.g. VMOVED) */
3137 if (!vp->specialStatus) {
3138 vp->specialStatus = VBUSY;
3150 #endif /* AFS_DEMAND_ATTACH_FS */
3153 * volume attachment helper function.
3155 * @param[out] ec error code
3156 * @param[in] volumeId volume ID of the attaching volume
3157 * @param[in] path full path to the volume header .vol file
3158 * @param[in] partp disk partition object for the attaching partition
3159 * @param[in] vp volume object; vp->hashid, vp->device, vp->partition,
3160 * vp->vnode_list, vp->rx_call_list, and V_attachCV (for
3161 * DAFS) should already be initialized
3162 * @param[in] isbusy 1 if vp->specialStatus should be set to VBUSY; that is,
3163 * if there is a volume operation running for this volume
3164 * that should set the volume to VBUSY during its run. 0
3165 * otherwise. (see VVolOpSetVBusy_r)
3166 * @param[in] mode attachment mode such as V_VOLUPD, V_DUMP, etc (see
3168 * @param[out] acheckedOut If we successfully checked-out the volume from
3169 * the fileserver (if we needed to), this is set
3170 * to 1, otherwise it is 0.
3172 * @return pointer to the semi-attached volume pointer
3173 * @retval NULL an error occurred (check value of *ec)
3174 * @retval vp volume successfully attaching
3176 * @pre no locks held
3178 * @post VOL_LOCK held
3181 attach2(Error * ec, VolumeId volumeId, char *path, struct DiskPartition64 *partp,
3182 Volume * vp, int isbusy, int mode, int *acheckedOut)
3184 /* have we read in the header successfully? */
3185 int read_header = 0;
3187 #ifdef AFS_DEMAND_ATTACH_FS
3188 /* should we FreeVolume(vp) instead of VCheckFree(vp) in the error
3192 /* in the case of an error, to what state should the volume be
3194 VolState error_state = VOL_STATE_ERROR;
3195 #endif /* AFS_DEMAND_ATTACH_FS */
3199 vp->vnodeIndex[vLarge].handle = NULL;
3200 vp->vnodeIndex[vSmall].handle = NULL;
3201 vp->diskDataHandle = NULL;
3202 vp->linkHandle = NULL;
3206 #ifdef AFS_DEMAND_ATTACH_FS
3207 attach_check_vop(ec, volumeId, partp, vp, acheckedOut);
3209 attach_volume_header(ec, vp, partp, mode, 0, acheckedOut);
3212 attach_volume_header(ec, vp, partp, mode, 0, acheckedOut);
3213 #endif /* !AFS_DEMAND_ATTACH_FS */
3215 if (*ec == VNOVOL) {
3216 /* if the volume doesn't exist, skip straight to 'error' so we don't
3217 * request a salvage */
3218 goto unlocked_error;
3224 /* ensure that we don't override specialStatus if it was set to
3225 * something else (e.g. VMOVED) */
3226 if (isbusy && !vp->specialStatus) {
3227 vp->specialStatus = VBUSY;
3229 vp->shuttingDown = 0;
3230 vp->goingOffline = 0;
3232 #ifdef AFS_DEMAND_ATTACH_FS
3233 vp->stats.last_attach = FT_ApproxTime();
3234 vp->stats.attaches++;
3238 IncUInt64(&VStats.attaches);
3239 vp->cacheCheck = ++VolumeCacheCheck;
3240 /* just in case this ever rolls over */
3241 if (!vp->cacheCheck)
3242 vp->cacheCheck = ++VolumeCacheCheck;
3245 #ifdef AFS_DEMAND_ATTACH_FS
3246 V_attachFlags(vp) |= VOL_HDR_LOADED;
3247 vp->stats.last_hdr_load = vp->stats.last_attach;
3248 #endif /* AFS_DEMAND_ATTACH_FS */
3252 struct IndexFileHeader iHead;
3255 * We just read in the diskstuff part of the header. If the detailed
3256 * volume stats area has not yet been initialized, we should bzero the
3257 * area and mark it as initialized.
3259 if (!(V_stat_initialized(vp))) {
3260 memset((V_stat_area(vp)), 0, VOL_STATS_BYTES);
3261 V_stat_initialized(vp) = 1;
3264 (void)ReadHeader(ec, vp->vnodeIndex[vSmall].handle,
3265 (char *)&iHead, sizeof(iHead),
3266 SMALLINDEXMAGIC, SMALLINDEXVERSION);
3269 Log("VAttachVolume: Error reading smallVnode vol header %s; error=%u\n", path, *ec);
3274 struct IndexFileHeader iHead;
3276 (void)ReadHeader(ec, vp->vnodeIndex[vLarge].handle,
3277 (char *)&iHead, sizeof(iHead),
3278 LARGEINDEXMAGIC, LARGEINDEXVERSION);
3281 Log("VAttachVolume: Error reading largeVnode vol header %s; error=%u\n", path, *ec);
3285 #ifdef AFS_NAMEI_ENV
3287 struct versionStamp stamp;
3289 (void)ReadHeader(ec, V_linkHandle(vp), (char *)&stamp,
3290 sizeof(stamp), LINKTABLEMAGIC, LINKTABLEVERSION);
3293 Log("VAttachVolume: Error reading namei vol header %s; error=%u\n", path, *ec);
3296 #endif /* AFS_NAMEI_ENV */
3298 #if defined(AFS_DEMAND_ATTACH_FS)
3299 if (*ec && ((*ec != VOFFLINE) || (V_attachState(vp) != VOL_STATE_UNATTACHED))) {
3301 if (!VCanScheduleSalvage()) {
3302 Log("VAttachVolume: Error attaching volume %s; volume needs salvage; error=%u\n", path, *ec);
3304 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_NO_OFFLINE);
3309 /* volume operation in progress */
3311 /* we have already transitioned the vp away from ATTACHING state, so we
3312 * can go right to the end of attach2, and we do not need to transition
3314 goto error_notbroken;
3316 #else /* AFS_DEMAND_ATTACH_FS */
3318 Log("VAttachVolume: Error attaching volume %s; volume needs salvage; error=%u\n", path, *ec);
3319 goto unlocked_error;
3321 #endif /* AFS_DEMAND_ATTACH_FS */
3323 if (V_needsSalvaged(vp)) {
3324 if (vp->specialStatus)
3325 vp->specialStatus = 0;
3327 #if defined(AFS_DEMAND_ATTACH_FS)
3328 if (!VCanScheduleSalvage()) {
3329 Log("VAttachVolume: volume salvage flag is ON for %s; volume needs salvage\n", path);
3331 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, VOL_SALVAGE_NO_OFFLINE);
3334 #else /* AFS_DEMAND_ATTACH_FS */
3336 #endif /* AFS_DEMAND_ATTACH_FS */
3342 vp->nextVnodeUnique = V_uniquifier(vp);
3344 if (VShouldCheckInUse(mode) && V_inUse(vp) && VolumeWriteable(vp)) {
3345 if (!V_needsSalvaged(vp)) {
3346 V_needsSalvaged(vp) = 1;
3347 VUpdateVolume_r(ec, vp, 0);
3349 #if defined(AFS_DEMAND_ATTACH_FS)
3350 if (!VCanScheduleSalvage()) {
3351 Log("VAttachVolume: volume %s needs to be salvaged; not attached.\n", path);
3353 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, VOL_SALVAGE_NO_OFFLINE);
3356 #else /* AFS_DEMAND_ATTACH_FS */
3357 Log("VAttachVolume: volume %s needs to be salvaged; not attached.\n", path);
3359 #endif /* AFS_DEMAND_ATTACH_FS */
3364 if (programType == fileServer && V_destroyMe(vp) == DESTROY_ME) {
3365 /* Only check destroyMe if we are the fileserver, since the
3366 * volserver et al sometimes need to work with volumes with
3367 * destroyMe set. Examples are 'temporary' volumes the
3368 * volserver creates, and when we create a volume (destroyMe
3369 * is set on creation; sometimes a separate volserver
3370 * transaction is created to clear destroyMe).
3373 #if defined(AFS_DEMAND_ATTACH_FS)
3374 /* schedule a salvage so the volume goes away on disk */
3375 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_NO_OFFLINE);
3376 VChangeState_r(vp, VOL_STATE_ERROR);
3379 #endif /* AFS_DEMAND_ATTACH_FS */
3380 Log("VAttachVolume: volume %s is junk; it should be destroyed at next salvage\n", path);
3385 vp->vnodeIndex[vSmall].bitmap = vp->vnodeIndex[vLarge].bitmap = NULL;
3386 #ifndef BITMAP_LATER
3387 if (programType == fileServer && VolumeWriteable(vp)) {
3389 for (i = 0; i < nVNODECLASSES; i++) {
3390 VGetBitmap_r(ec, vp, i);
3392 #ifdef AFS_DEMAND_ATTACH_FS
3393 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_NO_OFFLINE);
3395 #endif /* AFS_DEMAND_ATTACH_FS */
3396 Log("VAttachVolume: error getting bitmap for volume (%s)\n",
3402 #endif /* BITMAP_LATER */
3404 if (VInit >= 2 && V_needsCallback(vp)) {
3405 if (V_BreakVolumeCallbacks) {
3406 Log("VAttachVolume: Volume %lu was changed externally; breaking callbacks\n",
3407 afs_printable_uint32_lu(V_id(vp)));
3408 V_needsCallback(vp) = 0;
3410 (*V_BreakVolumeCallbacks) (V_id(vp));
3413 VUpdateVolume_r(ec, vp, 0);
3415 #ifdef FSSYNC_BUILD_CLIENT
3416 else if (VCanUseFSSYNC()) {
3417 afs_int32 fsync_code;
3419 V_needsCallback(vp) = 0;
3421 fsync_code = FSYNC_VolOp(V_id(vp), NULL, FSYNC_VOL_BREAKCBKS, FSYNC_WHATEVER, NULL);
3425 V_needsCallback(vp) = 1;
3426 Log("Error trying to tell the fileserver to break callbacks for "
3427 "changed volume %lu; error code %ld\n",
3428 afs_printable_uint32_lu(V_id(vp)),
3429 afs_printable_int32_ld(fsync_code));
3431 VUpdateVolume_r(ec, vp, 0);
3434 #endif /* FSSYNC_BUILD_CLIENT */
3437 Log("VAttachVolume: error %d clearing needsCallback on volume "
3438 "%lu; needs salvage\n", (int)*ec,
3439 afs_printable_uint32_lu(V_id(vp)));
3440 #ifdef AFS_DEMAND_ATTACH_FS
3441 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_NO_OFFLINE);
3443 #else /* !AFS_DEMAND_ATTACH_FS */
3445 #endif /* !AFS_DEMAND_ATTACh_FS */
3450 if (programType == fileServer) {
3451 if (vp->specialStatus)
3452 vp->specialStatus = 0;
3453 if (V_blessed(vp) && V_inService(vp) && !V_needsSalvaged(vp)) {
3454 V_inUse(vp) = fileServer;
3455 V_offlineMessage(vp)[0] = '\0';
3457 #ifdef AFS_DEMAND_ATTACH_FS
3458 /* check if the volume is actually usable. only do this for DAFS; for
3459 * non-DAFS, volumes that are not inService/blessed can still be
3460 * attached, even if clients cannot access them. this is relevant
3461 * because for non-DAFS, we try to attach the volume when e.g.
3462 * volserver gives us back then vol when its done with it, but
3463 * volserver may give us back a volume that is not inService/blessed. */
3467 /* Put the vol into PREATTACHED state, so if someone tries to
3468 * access it again, we try to attach, see that we're not blessed,
3469 * and give a VNOVOL error again. Putting it into UNATTACHED state
3470 * would result in a VOFFLINE error instead. */
3471 error_state = VOL_STATE_PREATTACHED;
3473 /* mimic e.g. GetVolume errors */
3474 if (!V_blessed(vp)) {
3475 Log("Volume %lu offline: not blessed\n", afs_printable_uint32_lu(V_id(vp)));
3476 FreeVolumeHeader(vp);
3477 } else if (!V_inService(vp)) {
3478 Log("Volume %lu offline: not in service\n", afs_printable_uint32_lu(V_id(vp)));
3479 FreeVolumeHeader(vp);
3481 Log("Volume %lu offline: needs salvage\n", afs_printable_uint32_lu(V_id(vp)));
3483 error_state = VOL_STATE_ERROR;
3484 /* see if we can recover */
3485 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, VOL_SALVAGE_NO_OFFLINE);
3490 #endif /* AFS_DEMAND_ATTACH_FS */
3492 #ifdef AFS_DEMAND_ATTACH_FS
3493 if ((mode != V_PEEK) && (mode != V_SECRETLY) && (mode != V_READONLY))
3494 V_inUse(vp) = programType;
3495 #endif /* AFS_DEMAND_ATTACH_FS */
3496 V_checkoutMode(vp) = mode;
3499 AddVolumeToHashTable(vp, V_id(vp));
3500 #ifdef AFS_DEMAND_ATTACH_FS
3501 if (VCanUnlockAttached() && (V_attachFlags(vp) & VOL_LOCKED)) {
3504 if ((programType != fileServer) ||
3505 (V_inUse(vp) == fileServer)) {
3506 AddVolumeToVByPList_r(vp);
3508 VChangeState_r(vp, VOL_STATE_ATTACHED);
3510 VChangeState_r(vp, VOL_STATE_UNATTACHED);
3519 #ifdef AFS_DEMAND_ATTACH_FS
3520 if (!VIsErrorState(V_attachState(vp))) {
3521 if (VIsErrorState(error_state)) {
3522 Log("attach2: forcing vol %" AFS_VOLID_FMT " to error state (state %u flags 0x%x ec %d)\n",
3523 afs_printable_VolumeId_lu(vp->hashid), V_attachState(vp),
3524 V_attachFlags(vp), *ec);
3526 VChangeState_r(vp, error_state);
3528 #endif /* AFS_DEMAND_ATTACH_FS */
3531 VReleaseVolumeHandles_r(vp);
3534 #ifdef AFS_DEMAND_ATTACH_FS
3536 if (VCheckSalvage(vp) == VCHECK_SALVAGE_FAIL) {
3537 /* The salvage could not be scheduled with the salvage server
3538 * due to a hard error. Reset the error code to prevent retry loops by
3540 if (*ec == VSALVAGING) {
3549 #else /* !AFS_DEMAND_ATTACH_FS */
3551 #endif /* !AFS_DEMAND_ATTACH_FS */
3555 /* Attach an existing volume.
3556 The volume also normally goes online at this time.
3557 An offline volume must be reattached to make it go online.
3561 VAttachVolume(Error * ec, VolumeId volumeId, int mode)
3565 retVal = VAttachVolume_r(ec, volumeId, mode);
3571 VAttachVolume_r(Error * ec, VolumeId volumeId, int mode)
3574 VGetVolumePath(ec, volumeId, &part, &name);
3578 vp = VGetVolume_r(&error, volumeId);
3580 opr_Assert(V_inUse(vp) == 0);
3581 VDetachVolume_r(ec, vp);
3585 return VAttachVolumeByName_r(ec, part, name, mode);
3588 /* Increment a reference count to a volume, sans context swaps. Requires
3589 * possibly reading the volume header in from the disk, since there's
3590 * an invariant in the volume package that nUsers>0 ==> vp->header is valid.
3592 * N.B. This call can fail if we can't read in the header!! In this case
3593 * we still guarantee we won't context swap, but the ref count won't be
3594 * incremented (otherwise we'd violate the invariant).
3596 /* NOTE: with the demand attach fileserver extensions, the global lock
3597 * is dropped within VHold */
3598 #ifdef AFS_DEMAND_ATTACH_FS
3600 VHold_r(Volume * vp)
3604 VCreateReservation_r(vp);
3605 VWaitExclusiveState_r(vp);
3607 LoadVolumeHeader(&error, vp);
3609 VCancelReservation_r(vp);
3613 VCancelReservation_r(vp);
3616 #else /* AFS_DEMAND_ATTACH_FS */
3618 VHold_r(Volume * vp)
3622 LoadVolumeHeader(&error, vp);
3628 #endif /* AFS_DEMAND_ATTACH_FS */
3630 /**** volume timeout-related stuff ****/
3632 #ifdef AFS_PTHREAD_ENV
3634 static struct timespec *shutdown_timeout;
3635 static pthread_once_t shutdown_timeout_once = PTHREAD_ONCE_INIT;
3638 VTimedOut(const struct timespec *ts)
3643 if (ts->tv_sec == 0) {
3644 /* short-circuit; this will have always timed out */
3648 code = gettimeofday(&tv, NULL);
3650 Log("Error %d from gettimeofday, assuming we have not timed out\n", errno);
3651 /* assume no timeout; failure mode is we just wait longer than normal
3652 * instead of returning errors when we shouldn't */
3656 if (tv.tv_sec < ts->tv_sec ||
3657 (tv.tv_sec == ts->tv_sec && tv.tv_usec*1000 < ts->tv_nsec)) {
3666 * Calculate an absolute timeout.
3668 * @param[out] ts A timeout that is "timeout" seconds from now, if we return
3669 * NULL, the memory is not touched
3670 * @param[in] timeout How long the timeout should be from now
3672 * @return timeout to use
3673 * @retval NULL no timeout; wait forever
3674 * @retval non-NULL the given value for "ts"
3678 static struct timespec *
3679 VCalcTimeout(struct timespec *ts, afs_int32 timeout)
3689 ts->tv_sec = ts->tv_nsec = 0;
3693 code = gettimeofday(&now, NULL);
3695 Log("Error %d from gettimeofday, falling back to 'forever' timeout\n", errno);
3699 ts->tv_sec = now.tv_sec + timeout;
3700 ts->tv_nsec = now.tv_usec * 1000;
3706 * Initialize the shutdown_timeout global.
3709 VShutdownTimeoutInit(void)
3711 struct timespec *ts;
3713 ts = malloc(sizeof(*ts));
3715 shutdown_timeout = VCalcTimeout(ts, vol_opts.offline_shutdown_timeout);
3717 if (!shutdown_timeout) {
3723 * Figure out the timeout that should be used for waiting for offline volumes.
3725 * @param[out] ats Storage space for a local timeout value if needed
3727 * @return The timeout value that should be used
3728 * @retval NULL No timeout; wait forever for offlining volumes
3729 * @retval non-NULL A pointer to the absolute time that should be used as
3730 * the deadline for waiting for offlining volumes.
3732 * @note If we return non-NULL, the pointer we return may or may not be the
3735 static const struct timespec *
3736 VOfflineTimeout(struct timespec *ats)
3738 if (vol_shutting_down) {
3739 opr_Verify(pthread_once(&shutdown_timeout_once,
3740 VShutdownTimeoutInit) == 0);
3741 return shutdown_timeout;
3743 return VCalcTimeout(ats, vol_opts.offline_timeout);
3747 #else /* AFS_PTHREAD_ENV */
3749 /* Waiting a certain amount of time for offlining volumes is not supported
3750 * for LWP due to a lack of primitives. So, we never time out */
3751 # define VTimedOut(x) (0)
3752 # define VOfflineTimeout(x) (NULL)
3754 #endif /* !AFS_PTHREAD_ENV */
3762 retVal = VHold_r(vp);
3769 VIsGoingOffline_r(struct Volume *vp)
3773 if (vp->goingOffline) {
3774 if (vp->specialStatus) {
3775 code = vp->specialStatus;
3776 } else if (V_inService(vp) == 0 || V_blessed(vp) == 0) {
3787 * Tell the caller if a volume is waiting to go offline.
3789 * @param[in] vp The volume we want to know about
3791 * @return volume status
3792 * @retval 0 volume is not waiting to go offline, go ahead and use it
3793 * @retval nonzero volume is waiting to offline, and give the returned code
3794 * as an error to anyone accessing the volume
3796 * @pre VOL_LOCK is NOT held
3797 * @pre caller holds a heavyweight reference on vp
3800 VIsGoingOffline(struct Volume *vp)
3805 code = VIsGoingOffline_r(vp);
3812 * Register an RX call with a volume.
3814 * @param[inout] ec Error code; if unset when passed in, may be set if
3815 * the volume starts going offline
3816 * @param[out] client_ec @see GetVolume
3817 * @param[in] vp Volume struct
3818 * @param[in] cbv VCallByVol struct containing the RX call to register
3820 * @pre VOL_LOCK held
3821 * @pre caller holds heavy ref on vp
3826 VRegisterCall_r(Error *ec, Error *client_ec, Volume *vp, struct VCallByVol *cbv)
3829 #ifdef AFS_DEMAND_ATTACH_FS
3831 /* just in case the volume started going offline after we got the
3832 * reference to it... otherwise, if the volume started going
3833 * offline right at the end of GetVolume(), we might race with the
3834 * RX call scanner, and return success and add our cbv to the
3835 * rx_call_list _after_ the scanner has scanned the list. */
3836 *ec = VIsGoingOffline_r(vp);
3842 while (V_attachState(vp) == VOL_STATE_SCANNING_RXCALLS) {
3843 VWaitStateChange_r(vp);
3845 #endif /* AFS_DEMAND_ATTACH_FS */
3847 queue_Prepend(&vp->rx_call_list, cbv);
3852 * Deregister an RX call with a volume.
3854 * @param[in] vp Volume struct
3855 * @param[in] cbv VCallByVol struct containing the RX call to deregister
3857 * @pre VOL_LOCK held
3858 * @pre caller holds heavy ref on vp
3863 VDeregisterCall_r(Volume *vp, struct VCallByVol *cbv)
3865 if (cbv && queue_IsOnQueue(cbv)) {
3866 #ifdef AFS_DEMAND_ATTACH_FS
3867 while (V_attachState(vp) == VOL_STATE_SCANNING_RXCALLS) {
3868 VWaitStateChange_r(vp);
3870 #endif /* AFS_DEMAND_ATTACH_FS */
3876 /***************************************************/
3877 /* get and put volume routines */
3878 /***************************************************/
3881 * put back a heavyweight reference to a volume object.
3883 * @param[in] vp volume object pointer
3885 * @pre VOL_LOCK held
3887 * @post heavyweight volume reference put back.
3888 * depending on state, volume may have been taken offline,
3889 * detached, salvaged, freed, etc.
3891 * @internal volume package internal use only
3894 VPutVolume_r(Volume * vp)
3896 opr_Verify(--vp->nUsers >= 0);
3897 if (vp->nUsers == 0) {
3899 ReleaseVolumeHeader(vp->header);
3900 #ifdef AFS_DEMAND_ATTACH_FS
3901 if (!VCheckDetach(vp)) {
3905 #else /* AFS_DEMAND_ATTACH_FS */
3907 #endif /* AFS_DEMAND_ATTACH_FS */
3912 VPutVolume(Volume * vp)
3920 * Puts a volume reference obtained with VGetVolumeWithCall.
3922 * @param[in] vp Volume struct
3923 * @param[in] cbv VCallByVol struct given to VGetVolumeWithCall, or NULL if none
3925 * @pre VOL_LOCK is NOT held
3928 VPutVolumeWithCall(Volume *vp, struct VCallByVol *cbv)
3931 VDeregisterCall_r(vp, cbv);
3936 /* Get a pointer to an attached volume. The pointer is returned regardless
3937 of whether or not the volume is in service or on/off line. An error
3938 code, however, is returned with an indication of the volume's status */
3940 VGetVolume(Error * ec, Error * client_ec, VolumeId volumeId)
3944 retVal = GetVolume(ec, client_ec, volumeId, NULL, 0);
3950 * Get a volume reference associated with an RX call.
3952 * @param[out] ec @see GetVolume
3953 * @param[out] client_ec @see GetVolume
3954 * @param[in] volumeId @see GetVolume
3955 * @param[in] ts How long to wait for going-offline volumes (absolute time).
3956 * If NULL, wait forever. If ts->tv_sec == 0, return immediately
3957 * with an error if the volume is going offline.
3958 * @param[in] cbv Contains an RX call to be associated with this volume
3959 * reference. This call may be interrupted if the volume is
3960 * requested to go offline while we hold a ref on it. Give NULL
3961 * to not associate an RX call with this reference.
3963 * @return @see GetVolume
3965 * @note for LWP builds, ts must be NULL
3967 * @note A reference obtained with this function MUST be put back with
3968 * VPutVolumeWithCall
3971 VGetVolumeWithCall(Error * ec, Error * client_ec, VolumeId volumeId,
3972 const struct timespec *ts, struct VCallByVol *cbv)
3976 retVal = GetVolume(ec, client_ec, volumeId, NULL, ts);
3977 VRegisterCall_r(ec, client_ec, retVal, cbv);
3983 VGetVolume_r(Error * ec, VolumeId volumeId)
3985 return GetVolume(ec, NULL, volumeId, NULL, NULL);
3988 /* try to get a volume we've previously looked up */
3989 /* for demand attach fs, caller MUST NOT hold a ref count on vp */
3991 VGetVolumeByVp_r(Error * ec, Volume * vp)
3993 return GetVolume(ec, NULL, vp->hashid, vp, NULL);
3997 * private interface for getting a volume handle
3999 * @param[out] ec error code (0 if no error)
4000 * @param[out] client_ec wire error code to be given to clients
4001 * @param[in] volumeId ID of the volume we want
4002 * @param[in] hint optional hint for hash lookups, or NULL
4003 * @param[in] timeout absolute deadline for waiting for the volume to go
4004 * offline, if it is going offline. NULL to wait forever.
4006 * @return a volume handle for the specified volume
4007 * @retval NULL an error occurred, or the volume is in such a state that
4008 * we cannot load a header or return any volume struct
4010 * @note for DAFS, caller must NOT hold a ref count on 'hint'
4012 * @note 'timeout' is only checked if the volume is actually going offline; so
4013 * if you pass timeout->tv_sec = 0, this will exhibit typical
4014 * nonblocking behavior.
4016 * @note for LWP builds, 'timeout' must be NULL
4019 GetVolume(Error * ec, Error * client_ec, VolumeId volumeId, Volume * hint,
4020 const struct timespec *timeout)
4023 /* pull this profiling/debugging code out of regular builds */
4025 #define VGET_CTR_INC(x) x++
4026 unsigned short V0 = 0, V1 = 0, V2 = 0, V3 = 0, V5 = 0, V6 =
4027 0, V7 = 0, V8 = 0, V9 = 0;
4028 unsigned short V10 = 0, V11 = 0, V12 = 0, V13 = 0, V14 = 0, V15 = 0;
4030 #define VGET_CTR_INC(x)
4032 #ifdef AFS_DEMAND_ATTACH_FS
4033 Volume *avp, * rvp = hint;
4037 * if VInit is zero, the volume package dynamic
4038 * data structures have not been initialized yet,
4039 * and we must immediately return an error
4045 *client_ec = VOFFLINE;
4050 #ifdef AFS_DEMAND_ATTACH_FS
4052 VCreateReservation_r(rvp);
4054 #endif /* AFS_DEMAND_ATTACH_FS */
4062 vp = VLookupVolume_r(ec, volumeId, vp);
4068 #ifdef AFS_DEMAND_ATTACH_FS
4069 if (rvp && (rvp != vp)) {
4070 /* break reservation on old vp */
4071 VCancelReservation_r(rvp);
4074 #endif /* AFS_DEMAND_ATTACH_FS */
4080 /* Until we have reached an initialization level of 2
4081 * we don't know whether this volume exists or not.
4082 * We can't sleep and retry later because before a volume
4083 * is attached, the caller tries to get it first. Just
4084 * return VOFFLINE and the caller can choose whether to
4085 * retry the command or not. */
4095 IncUInt64(&VStats.hdr_gets);
4097 #ifdef AFS_DEMAND_ATTACH_FS
4098 /* block if someone else is performing an exclusive op on this volume */
4101 VCreateReservation_r(rvp);
4103 VWaitExclusiveState_r(vp);
4105 /* short circuit with VNOVOL in the following circumstances:
4108 * - VOL_STATE_SHUTTING_DOWN
4110 if ((V_attachState(vp) == VOL_STATE_ERROR) ||
4111 (V_attachState(vp) == VOL_STATE_SHUTTING_DOWN)) {
4118 * short circuit with VOFFLINE for VOL_STATE_UNATTACHED/GOING_OFFLINE and
4119 * VNOVOL for VOL_STATE_DELETED
4121 if ((V_attachState(vp) == VOL_STATE_UNATTACHED) ||
4122 (V_attachState(vp) == VOL_STATE_GOING_OFFLINE) ||
4123 (V_attachState(vp) == VOL_STATE_DELETED)) {
4124 if (vp->specialStatus) {
4125 *ec = vp->specialStatus;
4126 } else if (V_attachState(vp) == VOL_STATE_DELETED) {
4135 /* allowable states:
4142 if (vp->salvage.requested) {
4143 VUpdateSalvagePriority_r(vp);
4146 if (V_attachState(vp) == VOL_STATE_PREATTACHED) {
4147 if (vp->specialStatus) {
4148 *ec = vp->specialStatus;
4152 avp = VAttachVolumeByVp_r(ec, vp, 0);
4155 /* VAttachVolumeByVp_r can return a pointer
4156 * != the vp passed to it under certain
4157 * conditions; make sure we don't leak
4158 * reservations if that happens */
4160 VCancelReservation_r(rvp);
4162 VCreateReservation_r(rvp);
4173 if (vp->specialStatus) {
4174 *ec = vp->specialStatus;
4179 if (vp->specialStatus) {
4180 *ec = vp->specialStatus;
4193 if (VIsSalvaging(vp) || (*ec == VSALVAGING)) {
4195 /* see CheckVnode() in afsfileprocs.c for an explanation
4196 * of this error code logic */
4197 afs_uint32 now = FT_ApproxTime();
4198 if ((vp->stats.last_salvage + (10 * 60)) >= now) {
4201 *client_ec = VRESTARTING;
4209 if (VIsErrorState(V_attachState(vp))) {
4210 /* make sure we don't take a vp in VOL_STATE_ERROR state and use
4211 * it, or transition it out of that state */
4220 * this test MUST happen after VAttachVolymeByVp, so we have no
4221 * conflicting vol op. (attach2 would have errored out if we had one;
4222 * specifically attach_check_vop must have detected a conflicting vop)
4224 opr_Assert(!vp->pending_vol_op || vp->pending_vol_op->vol_op_state == FSSYNC_VolOpRunningOnline);
4226 #endif /* AFS_DEMAND_ATTACH_FS */
4228 LoadVolumeHeader(ec, vp);
4231 /* Only log the error if it was a totally unexpected error. Simply
4232 * a missing inode is likely to be caused by the volume being deleted */
4233 if (errno != ENXIO || LogLevel)
4234 Log("Volume %" AFS_VOLID_FMT ": couldn't reread volume header\n",
4235 afs_printable_VolumeId_lu(vp->hashid));
4236 #ifdef AFS_DEMAND_ATTACH_FS
4237 if (VCanScheduleSalvage()) {
4238 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, 0 /*flags*/);
4243 #else /* AFS_DEMAND_ATTACH_FS */
4246 #endif /* AFS_DEMAND_ATTACH_FS */
4251 if (vp->shuttingDown) {
4258 if (programType == fileServer) {
4260 if (vp->goingOffline) {
4261 if (timeout && VTimedOut(timeout)) {
4262 /* we've timed out; don't wait for the vol */
4265 #ifdef AFS_DEMAND_ATTACH_FS
4266 /* wait for the volume to go offline */
4267 if (V_attachState(vp) == VOL_STATE_GOING_OFFLINE) {
4268 VTimedWaitStateChange_r(vp, timeout, NULL);
4270 #elif defined(AFS_PTHREAD_ENV)
4271 VOL_CV_TIMEDWAIT(&vol_put_volume_cond, timeout, NULL);
4272 #else /* AFS_PTHREAD_ENV */
4273 /* LWP has no timed wait, so the caller better not be
4275 opr_Assert(!timeout);
4276 LWP_WaitProcess(VPutVolume);
4277 #endif /* AFS_PTHREAD_ENV */
4281 if (vp->specialStatus) {
4283 *ec = vp->specialStatus;
4284 } else if (V_inService(vp) == 0 || V_blessed(vp) == 0) {
4287 } else if (V_inUse(vp) == 0 || vp->goingOffline) {
4298 #ifdef AFS_DEMAND_ATTACH_FS
4299 /* if no error, bump nUsers */
4302 VLRU_UpdateAccess_r(vp);
4305 VCancelReservation_r(rvp);
4308 if (client_ec && !*client_ec) {
4311 #else /* AFS_DEMAND_ATTACH_FS */
4312 /* if no error, bump nUsers */
4319 #endif /* AFS_DEMAND_ATTACH_FS */
4322 opr_Assert(vp || *ec);
4327 /***************************************************/
4328 /* Volume offline/detach routines */
4329 /***************************************************/
4331 /* caller MUST hold a heavyweight ref on vp */
4332 #ifdef AFS_DEMAND_ATTACH_FS
4334 VTakeOffline_r(Volume * vp)
4338 opr_Assert(vp->nUsers > 0);
4339 opr_Assert(programType == fileServer);
4341 VCreateReservation_r(vp);
4342 VWaitExclusiveState_r(vp);
4344 vp->goingOffline = 1;
4345 V_needsSalvaged(vp) = 1;
4347 VRequestSalvage_r(&error, vp, SALVSYNC_ERROR, 0);
4348 VCancelReservation_r(vp);
4350 #else /* AFS_DEMAND_ATTACH_FS */
4352 VTakeOffline_r(Volume * vp)
4354 opr_Assert(vp->nUsers > 0);
4355 opr_Assert(programType == fileServer);
4357 vp->goingOffline = 1;
4358 V_needsSalvaged(vp) = 1;
4360 #endif /* AFS_DEMAND_ATTACH_FS */
4363 VTakeOffline(Volume * vp)
4371 * force a volume offline.
4373 * @param[in] vp volume object pointer
4374 * @param[in] flags flags (see note below)
4376 * @note the flag VOL_FORCEOFF_NOUPDATE is a recursion control flag
4377 * used when VUpdateVolume_r needs to call VForceOffline_r
4378 * (which in turn would normally call VUpdateVolume_r)
4380 * @see VUpdateVolume_r
4382 * @pre VOL_LOCK must be held.
4383 * for DAFS, caller must hold ref.
4385 * @note for DAFS, it _is safe_ to call this function from an
4388 * @post needsSalvaged flag is set.
4389 * for DAFS, salvage is requested.
4390 * no further references to the volume through the volume
4391 * package will be honored.
4392 * all file descriptor and vnode caches are invalidated.
4394 * @warning this is a heavy-handed interface. it results in
4395 * a volume going offline regardless of the current
4396 * reference count state.
4398 * @internal volume package internal use only
4401 VForceOffline_r(Volume * vp, int flags)
4405 #ifdef AFS_DEMAND_ATTACH_FS
4406 VChangeState_r(vp, VOL_STATE_ERROR);
4411 strcpy(V_offlineMessage(vp),
4412 "Forced offline due to internal error: volume needs to be salvaged");
4413 Log("Volume %" AFS_VOLID_FMT " forced offline: it needs salvaging!\n", afs_printable_VolumeId_lu(V_id(vp)));
4416 vp->goingOffline = 0;
4417 V_needsSalvaged(vp) = 1;
4418 if (!(flags & VOL_FORCEOFF_NOUPDATE)) {
4419 VUpdateVolume_r(&error, vp, VOL_UPDATE_NOFORCEOFF);
4422 #ifdef AFS_DEMAND_ATTACH_FS
4423 VRequestSalvage_r(&error, vp, SALVSYNC_ERROR, 0 /*flags*/);
4424 #endif /* AFS_DEMAND_ATTACH_FS */
4426 #ifdef AFS_PTHREAD_ENV
4427 opr_cv_broadcast(&vol_put_volume_cond);
4428 #else /* AFS_PTHREAD_ENV */
4429 LWP_NoYieldSignal(VPutVolume);
4430 #endif /* AFS_PTHREAD_ENV */
4432 VReleaseVolumeHandles_r(vp);
4436 * force a volume offline.
4438 * @param[in] vp volume object pointer
4440 * @see VForceOffline_r
4443 VForceOffline(Volume * vp)
4446 VForceOffline_r(vp, 0);
4451 * Iterate over the RX calls associated with a volume, and interrupt them.
4453 * @param[in] vp The volume whose RX calls we want to scan
4455 * @pre VOL_LOCK held
4458 VScanCalls_r(struct Volume *vp)
4460 struct VCallByVol *cbv, *ncbv;
4462 #ifdef AFS_DEMAND_ATTACH_FS
4463 VolState state_save;
4466 if (queue_IsEmpty(&vp->rx_call_list))
4467 return; /* no calls to interrupt */
4468 if (!vol_opts.interrupt_rxcall)
4469 return; /* we have no function with which to interrupt calls */
4470 err = VIsGoingOffline_r(vp);
4472 return; /* we're not going offline anymore */
4474 #ifdef AFS_DEMAND_ATTACH_FS
4475 VWaitExclusiveState_r(vp);
4476 state_save = VChangeState_r(vp, VOL_STATE_SCANNING_RXCALLS);
4478 #endif /* AFS_DEMAND_ATTACH_FS */
4480 for(queue_Scan(&vp->rx_call_list, cbv, ncbv, VCallByVol)) {
4482 struct rx_peer *peer;
4484 peer = rx_PeerOf(rx_ConnectionOf(cbv->call));
4486 Log("Offlining volume %" AFS_VOLID_FMT " while client %s:%u is trying to read "
4487 "from it; kicking client off with error %ld\n",
4488 afs_printable_VolumeId_lu(vp->hashid),
4489 afs_inet_ntoa_r(rx_HostOf(peer), hoststr),
4490 (unsigned) ntohs(rx_PortOf(peer)),
4493 (*vol_opts.interrupt_rxcall) (cbv->call, err);
4496 #ifdef AFS_DEMAND_ATTACH_FS
4498 VChangeState_r(vp, state_save);
4499 #endif /* AFS_DEMAND_ATTACH_FS */
4502 #ifdef AFS_DEMAND_ATTACH_FS
4504 * Wait for a vp to go offline.
4506 * @param[out] ec 1 if a salvage on the volume has been requested and
4507 * salvok == 0, 0 otherwise
4508 * @param[in] vp The volume to wait for
4509 * @param[in] salvok If 0, we return immediately with *ec = 1 if the volume
4510 * has been requested to salvage. Otherwise we keep waiting
4511 * until the volume has gone offline.
4513 * @pre VOL_LOCK held
4514 * @pre caller holds a lightweight ref on vp
4519 VWaitForOfflineByVp_r(Error *ec, struct Volume *vp, int salvok)
4521 struct timespec timeout_ts;
4522 const struct timespec *ts;
4525 ts = VOfflineTimeout(&timeout_ts);
4529 while (!VIsOfflineState(V_attachState(vp)) && !timedout) {
4530 if (!salvok && vp->salvage.requested) {
4534 VTimedWaitStateChange_r(vp, ts, &timedout);
4537 /* we didn't time out, so the volume must be offline, so we're done */
4541 /* If we got here, we timed out waiting for the volume to go offline.
4542 * Kick off the accessing RX calls and wait again */
4546 while (!VIsOfflineState(V_attachState(vp))) {
4547 if (!salvok && vp->salvage.requested) {
4552 VWaitStateChange_r(vp);
4556 #else /* AFS_DEMAND_ATTACH_FS */
4559 * Wait for a volume to go offline.
4561 * @pre VOL_LOCK held
4563 * @note non-DAFS only (for DAFS, use @see WaitForOfflineByVp_r)
4566 VWaitForOffline_r(Error *ec, VolumeId volid)
4569 const struct timespec *ts;
4570 #ifdef AFS_PTHREAD_ENV
4571 struct timespec timeout_ts;
4574 ts = VOfflineTimeout(&timeout_ts);
4576 vp = GetVolume(ec, NULL, volid, NULL, ts);
4578 /* error occurred so bad that we can't even get a vp; we have no
4579 * information on the vol so we don't know whether to wait, so just
4583 if (!VIsGoingOffline_r(vp)) {
4584 /* volume is no longer going offline, so we're done */
4589 /* If we got here, we timed out waiting for the volume to go offline.
4590 * Kick off the accessing RX calls and wait again */
4596 vp = VGetVolume_r(ec, volid);
4598 /* In case it was reattached... */
4602 #endif /* !AFS_DEMAND_ATTACH_FS */
4604 /* The opposite of VAttachVolume. The volume header is written to disk, with
4605 the inUse bit turned off. A copy of the header is maintained in memory,
4606 however (which is why this is VOffline, not VDetach).
4609 VOffline_r(Volume * vp, char *message)
4612 #ifndef AFS_DEMAND_ATTACH_FS
4613 VolumeId vid = V_id(vp);
4616 opr_Assert(programType != volumeUtility && programType != volumeServer);
4621 if (V_offlineMessage(vp)[0] == '\0')
4622 strncpy(V_offlineMessage(vp), message, sizeof(V_offlineMessage(vp)));
4623 V_offlineMessage(vp)[sizeof(V_offlineMessage(vp)) - 1] = '\0';
4625 vp->goingOffline = 1;
4626 #ifdef AFS_DEMAND_ATTACH_FS
4627 VChangeState_r(vp, VOL_STATE_GOING_OFFLINE);
4628 VCreateReservation_r(vp);
4630 VWaitForOfflineByVp_r(&error, vp, 1);
4631 VCancelReservation_r(vp);
4632 #else /* AFS_DEMAND_ATTACH_FS */
4634 VWaitForOffline_r(&error, vid);
4635 #endif /* AFS_DEMAND_ATTACH_FS */
4638 #ifdef AFS_DEMAND_ATTACH_FS
4640 * Take a volume offline in order to perform a volume operation.
4642 * @param[inout] ec address in which to store error code
4643 * @param[in] vp volume object pointer
4644 * @param[in] message volume offline status message
4647 * - VOL_LOCK is held
4648 * - caller MUST hold a heavyweight ref on vp
4651 * - volume is taken offline
4652 * - if possible, volume operation is promoted to running state
4653 * - on failure, *ec is set to nonzero
4655 * @note Although this function does not return any value, it may
4656 * still fail to promote our pending volume operation to
4657 * a running state. Any caller MUST check the value of *ec,
4658 * and MUST NOT blindly assume success.
4660 * @warning if the caller does not hold a lightweight ref on vp,
4661 * then it MUST NOT reference vp after this function
4662 * returns to the caller.
4664 * @internal volume package internal use only
4667 VOfflineForVolOp_r(Error *ec, Volume *vp, char *message)
4670 opr_Assert(vp->pending_vol_op);
4676 if (V_offlineMessage(vp)[0] == '\0')
4677 strncpy(V_offlineMessage(vp), message, sizeof(V_offlineMessage(vp)));
4678 V_offlineMessage(vp)[sizeof(V_offlineMessage(vp)) - 1] = '\0';
4680 vp->goingOffline = 1;
4681 VChangeState_r(vp, VOL_STATE_GOING_OFFLINE);
4682 VCreateReservation_r(vp);
4685 if (vp->pending_vol_op->com.programType != salvageServer) {
4686 /* do not give corrupted volumes to the volserver */
4691 VWaitForOfflineByVp_r(ec, vp, salvok);
4693 VCancelReservation_r(vp);
4695 #endif /* AFS_DEMAND_ATTACH_FS */
4698 VOffline(Volume * vp, char *message)
4701 VOffline_r(vp, message);
4705 /* This gets used for the most part by utility routines that don't want
4706 * to keep all the volume headers around. Generally, the file server won't
4707 * call this routine, because then the offline message in the volume header
4708 * (or other information) won't be available to clients. For NAMEI, also
4709 * close the file handles. However, the fileserver does call this during
4710 * an attach following a volume operation.
4713 VDetachVolume_r(Error * ec, Volume * vp)
4715 #ifdef FSSYNC_BUILD_CLIENT
4717 struct DiskPartition64 *tpartp;
4718 int notifyServer = 0;
4719 int useDone = FSYNC_VOL_ON;
4721 if (VCanUseFSSYNC()) {
4722 notifyServer = vp->needsPutBack;
4723 if (V_destroyMe(vp) == DESTROY_ME)
4724 useDone = FSYNC_VOL_LEAVE_OFF;
4725 # ifdef AFS_DEMAND_ATTACH_FS
4726 else if (!V_blessed(vp) || !V_inService(vp))
4727 useDone = FSYNC_VOL_LEAVE_OFF;
4730 # ifdef AFS_DEMAND_ATTACH_FS
4731 if (V_needsSalvaged(vp)) {
4733 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, 0);
4736 tpartp = vp->partition;
4738 #endif /* FSSYNC_BUILD_CLIENT */
4740 *ec = 0; /* always "succeeds" */
4741 DeleteVolumeFromHashTable(vp);
4742 vp->shuttingDown = 1;
4743 #ifdef AFS_DEMAND_ATTACH_FS
4744 DeleteVolumeFromVByPList_r(vp);
4746 VChangeState_r(vp, VOL_STATE_SHUTTING_DOWN);
4748 if (programType != fileServer)
4750 #endif /* AFS_DEMAND_ATTACH_FS */
4752 /* Will be detached sometime in the future--this is OK since volume is offline */
4754 /* XXX the following code should really be moved to VCheckDetach() since the volume
4755 * is not technically detached until the refcounts reach zero
4757 #ifdef FSSYNC_BUILD_CLIENT
4758 if (VCanUseFSSYNC() && notifyServer) {
4759 if (notifyServer == VOL_PUTBACK_DELETE) {
4760 /* Only send FSYNC_VOL_DONE if the volume was actually deleted.
4761 * volserver code will set needsPutBack to VOL_PUTBACK_DELETE
4762 * to signify a deleted volume. */
4763 useDone = FSYNC_VOL_DONE;
4766 * Note: The server is not notified in the case of a bogus volume
4767 * explicitly to make it possible to create a volume, do a partial
4768 * restore, then abort the operation without ever putting the volume
4769 * online. This is essential in the case of a volume move operation
4770 * between two partitions on the same server. In that case, there
4771 * would be two instances of the same volume, one of them bogus,
4772 * which the file server would attempt to put on line
4774 FSYNC_VolOp(volume, tpartp->name, useDone, 0, NULL);
4775 /* XXX this code path is only hit by volume utilities, thus
4776 * V_BreakVolumeCallbacks will always be NULL. if we really
4777 * want to break callbacks in this path we need to use FSYNC_VolOp() */
4779 /* Dettaching it so break all callbacks on it */
4780 if (V_BreakVolumeCallbacks) {
4781 Log("volume %u detached; breaking all call backs\n", volume);
4782 (*V_BreakVolumeCallbacks) (volume);
4786 #endif /* FSSYNC_BUILD_CLIENT */
4790 VDetachVolume(Error * ec, Volume * vp)
4793 VDetachVolume_r(ec, vp);
4798 /***************************************************/
4799 /* Volume fd/inode handle closing routines */
4800 /***************************************************/
4802 /* For VDetachVolume, we close all cached file descriptors, but keep
4803 * the Inode handles in case we need to read from a busy volume.
4805 /* for demand attach, caller MUST hold ref count on vp */
4807 VCloseVolumeHandles_r(Volume * vp)
4809 #ifdef AFS_DEMAND_ATTACH_FS
4810 VolState state_save;
4812 state_save = VChangeState_r(vp, VOL_STATE_OFFLINING);
4817 DFlushVolume(vp->hashid);
4819 #ifdef AFS_DEMAND_ATTACH_FS
4823 /* DAFS: VCloseVnodeFiles_r drops the glock internally */
4824 VCloseVnodeFiles_r(vp);
4826 #ifdef AFS_DEMAND_ATTACH_FS
4830 /* Too time consuming and unnecessary for the volserver */
4831 if (programType == fileServer) {
4832 IH_CONDSYNC(vp->vnodeIndex[vLarge].handle);
4833 IH_CONDSYNC(vp->vnodeIndex[vSmall].handle);
4834 IH_CONDSYNC(vp->diskDataHandle);
4835 #ifdef AFS_NAMEI_ENV
4836 IH_CONDSYNC(vp->linkHandle);
4837 #endif /* AFS_NAMEI_ENV */
4840 IH_REALLYCLOSE(vp->vnodeIndex[vLarge].handle);
4841 IH_REALLYCLOSE(vp->vnodeIndex[vSmall].handle);
4842 IH_REALLYCLOSE(vp->diskDataHandle);
4843 IH_REALLYCLOSE(vp->linkHandle);
4845 #ifdef AFS_DEMAND_ATTACH_FS
4846 if ((V_attachFlags(vp) & VOL_LOCKED)) {
4851 VChangeState_r(vp, state_save);
4855 /* For both VForceOffline and VOffline, we close all relevant handles.
4856 * For VOffline, if we re-attach the volume, the files may possible be
4857 * different than before.
4859 /* for demand attach, caller MUST hold a ref count on vp */
4861 VReleaseVolumeHandles_r(Volume * vp)
4863 #ifdef AFS_DEMAND_ATTACH_FS
4864 VolState state_save;
4866 state_save = VChangeState_r(vp, VOL_STATE_DETACHING);
4871 DFlushVolume(vp->hashid);
4873 #ifdef AFS_DEMAND_ATTACH_FS
4877 VReleaseVnodeFiles_r(vp); /* DAFS: releases the glock internally */
4879 #ifdef AFS_DEMAND_ATTACH_FS
4883 /* Too time consuming and unnecessary for the volserver */
4884 if (programType == fileServer) {
4885 IH_CONDSYNC(vp->vnodeIndex[vLarge].handle);
4886 IH_CONDSYNC(vp->vnodeIndex[vSmall].handle);
4887 IH_CONDSYNC(vp->diskDataHandle);
4888 #ifdef AFS_NAMEI_ENV
4889 IH_CONDSYNC(vp->linkHandle);
4890 #endif /* AFS_NAMEI_ENV */
4893 IH_RELEASE(vp->vnodeIndex[vLarge].handle);
4894 IH_RELEASE(vp->vnodeIndex[vSmall].handle);
4895 IH_RELEASE(vp->diskDataHandle);
4896 IH_RELEASE(vp->linkHandle);
4898 #ifdef AFS_DEMAND_ATTACH_FS
4899 if ((V_attachFlags(vp) & VOL_LOCKED)) {
4904 VChangeState_r(vp, state_save);
4909 /***************************************************/
4910 /* Volume write and fsync routines */
4911 /***************************************************/
4914 VUpdateVolume_r(Error * ec, Volume * vp, int flags)
4916 #ifdef AFS_DEMAND_ATTACH_FS
4917 VolState state_save;
4919 if (flags & VOL_UPDATE_WAIT) {
4920 VCreateReservation_r(vp);
4921 VWaitExclusiveState_r(vp);
4926 if (programType == fileServer)
4928 (V_inUse(vp) ? V_nextVnodeUnique(vp) +
4929 200 : V_nextVnodeUnique(vp));
4931 #ifdef AFS_DEMAND_ATTACH_FS
4932 state_save = VChangeState_r(vp, VOL_STATE_UPDATING);
4936 WriteVolumeHeader_r(ec, vp);
4938 #ifdef AFS_DEMAND_ATTACH_FS
4940 VChangeState_r(vp, state_save);
4941 if (flags & VOL_UPDATE_WAIT) {
4942 VCancelReservation_r(vp);
4947 Log("VUpdateVolume: error updating volume header, volume %" AFS_VOLID_FMT " (%s)\n",
4948 afs_printable_VolumeId_lu(V_id(vp)), V_name(vp));
4949 /* try to update on-disk header,
4950 * while preventing infinite recursion */
4951 if (!(flags & VOL_UPDATE_NOFORCEOFF)) {
4952 VForceOffline_r(vp, VOL_FORCEOFF_NOUPDATE);
4958 VUpdateVolume(Error * ec, Volume * vp)
4961 VUpdateVolume_r(ec, vp, VOL_UPDATE_WAIT);
4966 VSyncVolume_r(Error * ec, Volume * vp, int flags)
4970 #ifdef AFS_DEMAND_ATTACH_FS
4971 VolState state_save;
4974 if (flags & VOL_SYNC_WAIT) {
4975 VUpdateVolume_r(ec, vp, VOL_UPDATE_WAIT);
4977 VUpdateVolume_r(ec, vp, 0);
4980 #ifdef AFS_DEMAND_ATTACH_FS
4981 state_save = VChangeState_r(vp, VOL_STATE_UPDATING);
4984 fdP = IH_OPEN(V_diskDataHandle(vp));
4985 opr_Assert(fdP != NULL);
4986 code = FDH_SYNC(fdP);
4987 opr_Assert(code == 0);
4989 #ifdef AFS_DEMAND_ATTACH_FS
4991 VChangeState_r(vp, state_save);
4997 VSyncVolume(Error * ec, Volume * vp)
5000 VSyncVolume_r(ec, vp, VOL_SYNC_WAIT);
5005 /***************************************************/
5006 /* Volume dealloaction routines */
5007 /***************************************************/
5009 #ifdef AFS_DEMAND_ATTACH_FS
5011 FreeVolume(Volume * vp)
5013 /* free the heap space, iff it's safe.
5014 * otherwise, pull it out of the hash table, so it
5015 * will get deallocated when all refs to it go away */
5016 if (!VCheckFree(vp)) {
5017 DeleteVolumeFromHashTable(vp);
5018 DeleteVolumeFromVByPList_r(vp);
5020 /* make sure we invalidate the header cache entry */
5021 FreeVolumeHeader(vp);
5024 #endif /* AFS_DEMAND_ATTACH_FS */
5027 ReallyFreeVolume(Volume * vp)
5032 #ifdef AFS_DEMAND_ATTACH_FS
5034 VChangeState_r(vp, VOL_STATE_FREED);
5035 if (vp->pending_vol_op)
5036 free(vp->pending_vol_op);
5037 #endif /* AFS_DEMAND_ATTACH_FS */
5038 for (i = 0; i < nVNODECLASSES; i++)
5039 if (vp->vnodeIndex[i].bitmap)
5040 free(vp->vnodeIndex[i].bitmap);
5041 FreeVolumeHeader(vp);
5042 #ifndef AFS_DEMAND_ATTACH_FS
5043 DeleteVolumeFromHashTable(vp);
5044 #endif /* AFS_DEMAND_ATTACH_FS */
5048 /* check to see if we should shutdown this volume
5049 * returns 1 if volume was freed, 0 otherwise */
5050 #ifdef AFS_DEMAND_ATTACH_FS
5052 VCheckDetach(Volume * vp)
5057 if (vp->nUsers || vp->nWaiters)
5060 if (vp->shuttingDown) {
5062 if ((programType != fileServer) &&
5063 (V_inUse(vp) == programType) &&
5064 ((V_checkoutMode(vp) == V_VOLUPD) ||
5065 (V_checkoutMode(vp) == V_SECRETLY) ||
5066 ((V_checkoutMode(vp) == V_CLONE) &&
5067 (VolumeWriteable(vp))))) {
5069 VUpdateVolume_r(&ec, vp, VOL_UPDATE_NOFORCEOFF);
5071 Log("VCheckDetach: volume header update for volume %" AFS_VOLID_FMT " "
5072 "failed with errno %d\n", afs_printable_VolumeId_lu(vp->hashid), errno);
5075 VReleaseVolumeHandles_r(vp);
5077 ReallyFreeVolume(vp);
5078 if (programType == fileServer) {
5079 opr_cv_broadcast(&vol_put_volume_cond);
5084 #else /* AFS_DEMAND_ATTACH_FS */
5086 VCheckDetach(Volume * vp)
5094 if (vp->shuttingDown) {
5096 if ((programType != fileServer) &&
5097 (V_inUse(vp) == programType) &&
5098 ((V_checkoutMode(vp) == V_VOLUPD) ||
5099 (V_checkoutMode(vp) == V_SECRETLY) ||
5100 ((V_checkoutMode(vp) == V_CLONE) &&
5101 (VolumeWriteable(vp))))) {
5103 VUpdateVolume_r(&ec, vp, VOL_UPDATE_NOFORCEOFF);
5105 Log("VCheckDetach: volume header update for volume %" AFS_VOLID_FMT " failed with errno %d\n",
5106 afs_printable_VolumeId_lu(vp->hashid), errno);
5109 VReleaseVolumeHandles_r(vp);
5110 ReallyFreeVolume(vp);
5111 if (programType == fileServer) {
5112 #if defined(AFS_PTHREAD_ENV)
5113 opr_cv_broadcast(&vol_put_volume_cond);
5114 #else /* AFS_PTHREAD_ENV */
5115 LWP_NoYieldSignal(VPutVolume);
5116 #endif /* AFS_PTHREAD_ENV */
5121 #endif /* AFS_DEMAND_ATTACH_FS */
5123 /* check to see if we should offline this volume
5124 * return 1 if volume went offline, 0 otherwise */
5125 #ifdef AFS_DEMAND_ATTACH_FS
5127 VCheckOffline(Volume * vp)
5131 if (vp->goingOffline && !vp->nUsers) {
5133 opr_Assert(programType == fileServer);
5134 opr_Assert((V_attachState(vp) != VOL_STATE_ATTACHED) &&
5135 (V_attachState(vp) != VOL_STATE_FREED) &&
5136 (V_attachState(vp) != VOL_STATE_PREATTACHED) &&
5137 (V_attachState(vp) != VOL_STATE_UNATTACHED) &&
5138 (V_attachState(vp) != VOL_STATE_DELETED));
5142 * VOL_STATE_GOING_OFFLINE
5143 * VOL_STATE_SHUTTING_DOWN
5144 * VIsErrorState(V_attachState(vp))
5145 * VIsExclusiveState(V_attachState(vp))
5148 VCreateReservation_r(vp);
5149 VChangeState_r(vp, VOL_STATE_OFFLINING);
5152 /* must clear the goingOffline flag before we drop the glock */
5153 vp->goingOffline = 0;
5158 /* perform async operations */
5159 VUpdateVolume_r(&error, vp, 0);
5160 VCloseVolumeHandles_r(vp);
5163 if (V_offlineMessage(vp)[0]) {
5164 Log("VOffline: Volume %lu (%s) is now offline (%s)\n",
5165 afs_printable_uint32_lu(V_id(vp)), V_name(vp),
5166 V_offlineMessage(vp));
5168 Log("VOffline: Volume %lu (%s) is now offline\n",
5169 afs_printable_uint32_lu(V_id(vp)), V_name(vp));
5173 /* invalidate the volume header cache entry */
5174 FreeVolumeHeader(vp);
5176 /* if nothing changed state to error or salvaging,
5177 * drop state to unattached */
5178 if (!VIsErrorState(V_attachState(vp))) {
5179 VChangeState_r(vp, VOL_STATE_UNATTACHED);
5181 VCancelReservation_r(vp);
5182 /* no usage of vp is safe beyond this point */
5186 #else /* AFS_DEMAND_ATTACH_FS */
5188 VCheckOffline(Volume * vp)
5192 if (vp->goingOffline && !vp->nUsers) {
5194 opr_Assert(programType == fileServer);
5197 vp->goingOffline = 0;
5199 VUpdateVolume_r(&error, vp, 0);
5200 VCloseVolumeHandles_r(vp);
5202 if (V_offlineMessage(vp)[0]) {
5203 Log("VOffline: Volume %lu (%s) is now offline (%s)\n",
5204 afs_printable_uint32_lu(V_id(vp)), V_name(vp),
5205 V_offlineMessage(vp));
5207 Log("VOffline: Volume %lu (%s) is now offline\n",
5208 afs_printable_uint32_lu(V_id(vp)), V_name(vp));
5211 FreeVolumeHeader(vp);
5212 #ifdef AFS_PTHREAD_ENV
5213 opr_cv_broadcast(&vol_put_volume_cond);
5214 #else /* AFS_PTHREAD_ENV */
5215 LWP_NoYieldSignal(VPutVolume);
5216 #endif /* AFS_PTHREAD_ENV */
5220 #endif /* AFS_DEMAND_ATTACH_FS */
5222 /***************************************************/
5223 /* demand attach fs ref counting routines */
5224 /***************************************************/
5226 #ifdef AFS_DEMAND_ATTACH_FS
5227 /* the following two functions handle reference counting for
5228 * asynchronous operations on volume structs.
5230 * their purpose is to prevent a VDetachVolume or VShutdown
5231 * from free()ing the Volume struct during an async i/o op */
5233 /* register with the async volume op ref counter */
5234 /* VCreateReservation_r moved into inline code header because it
5235 * is now needed in vnode.c -- tkeiser 11/20/2007
5239 * decrement volume-package internal refcount.
5241 * @param vp volume object pointer
5243 * @internal volume package internal use only
5246 * @arg VOL_LOCK is held
5247 * @arg lightweight refcount held
5249 * @post volume waiters refcount is decremented; volume may
5250 * have been deallocated/shutdown/offlined/salvaged/
5251 * whatever during the process
5253 * @warning once you have tossed your last reference (you can acquire
5254 * lightweight refs recursively) it is NOT SAFE to reference
5255 * a volume object pointer ever again
5257 * @see VCreateReservation_r
5259 * @note DEMAND_ATTACH_FS only
5262 VCancelReservation_r(Volume * vp)
5264 opr_Verify(--vp->nWaiters >= 0);
5265 if (vp->nWaiters == 0) {
5267 if (!VCheckDetach(vp)) {
5274 /* check to see if we should free this volume now
5275 * return 1 if volume was freed, 0 otherwise */
5277 VCheckFree(Volume * vp)
5280 if ((vp->nUsers == 0) &&
5281 (vp->nWaiters == 0) &&
5282 !(V_attachFlags(vp) & (VOL_IN_HASH |
5286 ReallyFreeVolume(vp);
5291 #endif /* AFS_DEMAND_ATTACH_FS */
5294 /***************************************************/
5295 /* online volume operations routines */
5296 /***************************************************/
5298 #ifdef AFS_DEMAND_ATTACH_FS
5300 * register a volume operation on a given volume.
5302 * @param[in] vp volume object
5303 * @param[in] vopinfo volume operation info object
5305 * @pre VOL_LOCK is held
5307 * @post volume operation info object attached to volume object.
5308 * volume operation statistics updated.
5310 * @note by "attached" we mean a copy of the passed in object is made
5312 * @internal volume package internal use only
5315 VRegisterVolOp_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
5317 FSSYNC_VolOp_info * info;
5319 /* attach a vol op info node to the volume struct */
5320 info = malloc(sizeof(FSSYNC_VolOp_info));
5321 opr_Assert(info != NULL);
5322 memcpy(info, vopinfo, sizeof(FSSYNC_VolOp_info));
5323 vp->pending_vol_op = info;
5326 vp->stats.last_vol_op = FT_ApproxTime();
5327 vp->stats.vol_ops++;
5328 IncUInt64(&VStats.vol_ops);
5334 * deregister the volume operation attached to this volume.
5336 * @param[in] vp volume object pointer
5338 * @pre VOL_LOCK is held
5340 * @post the volume operation info object is detached from the volume object
5342 * @internal volume package internal use only
5345 VDeregisterVolOp_r(Volume * vp)
5347 if (vp->pending_vol_op) {
5348 free(vp->pending_vol_op);
5349 vp->pending_vol_op = NULL;
5353 #endif /* AFS_DEMAND_ATTACH_FS */
5356 * determine whether it is safe to leave a volume online during
5357 * the volume operation described by the vopinfo object.
5359 * @param[in] vp volume object
5360 * @param[in] vopinfo volume operation info object
5362 * @return whether it is safe to leave volume online
5363 * @retval 0 it is NOT SAFE to leave the volume online
5364 * @retval 1 it is safe to leave the volume online during the operation
5367 * @arg VOL_LOCK is held
5368 * @arg disk header attached to vp (heavyweight ref on vp will guarantee
5369 * this condition is met)
5371 * @internal volume package internal use only
5374 VVolOpLeaveOnline_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
5376 return (vopinfo->vol_op_state == FSSYNC_VolOpRunningOnline ||
5377 (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
5378 (vopinfo->com.reason == V_READONLY ||
5379 (!VolumeWriteable(vp) &&
5380 (vopinfo->com.reason == V_CLONE ||
5381 vopinfo->com.reason == V_DUMP)))));
5385 * same as VVolOpLeaveOnline_r, but does not require a volume with an attached
5388 * @param[in] vp volume object
5389 * @param[in] vopinfo volume operation info object
5391 * @return whether it is safe to leave volume online
5392 * @retval 0 it is NOT SAFE to leave the volume online
5393 * @retval 1 it is safe to leave the volume online during the operation
5394 * @retval -1 unsure; volume header is required in order to know whether or
5395 * not is is safe to leave the volume online
5397 * @pre VOL_LOCK is held
5399 * @internal volume package internal use only
5402 VVolOpLeaveOnlineNoHeader_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
5404 /* follow the logic in VVolOpLeaveOnline_r; this is the same, except
5405 * assume that we don't know VolumeWriteable; return -1 if the answer
5406 * depends on VolumeWriteable */
5408 if (vopinfo->vol_op_state == FSSYNC_VolOpRunningOnline) {
5411 if (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
5412 vopinfo->com.reason == V_READONLY) {
5416 if (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
5417 (vopinfo->com.reason == V_CLONE ||
5418 vopinfo->com.reason == V_DUMP)) {
5420 /* must know VolumeWriteable */
5427 * determine whether VBUSY should be set during this volume operation.
5429 * @param[in] vp volume object
5430 * @param[in] vopinfo volume operation info object
5432 * @return whether VBUSY should be set
5433 * @retval 0 VBUSY does NOT need to be set
5434 * @retval 1 VBUSY SHOULD be set
5436 * @pre VOL_LOCK is held
5438 * @internal volume package internal use only
5441 VVolOpSetVBusy_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
5443 return ((vopinfo->com.command == FSYNC_VOL_OFF &&
5444 vopinfo->com.reason == FSYNC_SALVAGE) ||
5445 (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
5446 (vopinfo->com.reason == V_CLONE ||
5447 vopinfo->com.reason == V_DUMP)));
5451 /***************************************************/
5452 /* online salvager routines */
5453 /***************************************************/
5454 #if defined(AFS_DEMAND_ATTACH_FS)
5457 * offline a volume to let it be salvaged.
5459 * @param[in] vp Volume to offline
5461 * @return whether we offlined the volume successfully
5462 * @retval 0 volume was not offlined
5463 * @retval 1 volume is now offline
5465 * @note This is similar to VCheckOffline, but slightly different. We do not
5466 * deal with vp->goingOffline, and we try to avoid touching the volume
5467 * header except just to set needsSalvaged
5469 * @pre VOL_LOCK held
5470 * @pre vp->nUsers == 0
5471 * @pre V_attachState(vp) == VOL_STATE_SALVAGE_REQ
5474 VOfflineForSalvage_r(struct Volume *vp)
5478 VCreateReservation_r(vp);
5479 VWaitExclusiveState_r(vp);
5481 if (vp->nUsers || V_attachState(vp) == VOL_STATE_SALVAGING) {
5482 /* Someone's using the volume, or someone got to scheduling the salvage
5483 * before us. I don't think either of these should be possible, as we
5484 * should gain no new heavyweight references while we're trying to
5485 * salvage, but just to be sure... */
5486 VCancelReservation_r(vp);
5490 VChangeState_r(vp, VOL_STATE_OFFLINING);
5494 V_needsSalvaged(vp) = 1;
5495 /* ignore error; updating needsSalvaged is just best effort */
5496 VUpdateVolume_r(&error, vp, VOL_UPDATE_NOFORCEOFF);
5498 VCloseVolumeHandles_r(vp);
5500 FreeVolumeHeader(vp);
5502 /* volume has been effectively offlined; we can mark it in the SALVAGING
5503 * state now, which lets FSSYNC give it away */
5504 VChangeState_r(vp, VOL_STATE_SALVAGING);
5506 VCancelReservation_r(vp);
5512 * check whether a salvage needs to be performed on this volume.
5514 * @param[in] vp pointer to volume object
5516 * @return status code
5517 * @retval VCHECK_SALVAGE_OK (0) no pending salvage
5518 * @retval VCHECK_SALVAGE_SCHEDULED (1) salvage has been scheduled
5519 * @retval VCHECK_SALVAGE_ASYNC (2) salvage being scheduled
5520 * @retval VCHECK_SALVAGE_DENIED (3) salvage not scheduled; denied
5521 * @retval VCHECK_SALVAGE_FAIL (4) salvage not scheduled; failed
5523 * @pre VOL_LOCK is held
5525 * @post if salvage request flag is set and nUsers and nWaiters are zero,
5526 * then a salvage will be requested
5528 * @note this is one of the event handlers called by VCancelReservation_r
5530 * @note the caller must check if the volume needs to be freed after calling
5531 * this; the volume may not have any references or be on any lists after
5532 * we return, and we do not free it
5534 * @see VCancelReservation_r
5536 * @internal volume package internal use only.
5539 VCheckSalvage(Volume * vp)
5541 int ret = VCHECK_SALVAGE_OK;
5543 #if defined(SALVSYNC_BUILD_CLIENT) || defined(FSSYNC_BUILD_CLIENT)
5544 if (!vp->salvage.requested) {
5545 return VCHECK_SALVAGE_OK;
5548 return VCHECK_SALVAGE_ASYNC;
5551 /* prevent recursion; some of the code below creates and removes
5552 * lightweight refs, which can call VCheckSalvage */
5553 if (vp->salvage.scheduling) {
5554 return VCHECK_SALVAGE_ASYNC;
5556 vp->salvage.scheduling = 1;
5558 if (V_attachState(vp) == VOL_STATE_SALVAGE_REQ) {
5559 if (!VOfflineForSalvage_r(vp)) {
5560 vp->salvage.scheduling = 0;
5561 return VCHECK_SALVAGE_FAIL;
5565 if (vp->salvage.requested) {
5566 ret = VScheduleSalvage_r(vp);
5568 vp->salvage.scheduling = 0;
5569 #endif /* SALVSYNC_BUILD_CLIENT || FSSYNC_BUILD_CLIENT */
5574 * request volume salvage.
5576 * @param[out] ec computed client error code
5577 * @param[in] vp volume object pointer
5578 * @param[in] reason reason code (passed to salvageserver via SALVSYNC)
5579 * @param[in] flags see flags note below
5582 * VOL_SALVAGE_NO_OFFLINE do not need to wait to offline the volume; it has
5583 * not been fully attached
5585 * @pre VOL_LOCK is held.
5587 * @post volume state is changed.
5588 * for fileserver, salvage will be requested once refcount reaches zero.
5590 * @return operation status code
5591 * @retval 0 volume salvage will occur
5592 * @retval 1 volume salvage could not be scheduled
5596 * @note in the fileserver, this call does not synchronously schedule a volume
5597 * salvage. rather, it sets volume state so that when volume refcounts
5598 * reach zero, a volume salvage will occur. by "refcounts", we mean both
5599 * nUsers and nWaiters must be zero.
5601 * @internal volume package internal use only.
5604 VRequestSalvage_r(Error * ec, Volume * vp, int reason, int flags)
5608 * for DAFS volume utilities that are not supposed to schedule salvages,
5609 * just transition to error state instead
5611 if (!VCanScheduleSalvage()) {
5612 VChangeState_r(vp, VOL_STATE_ERROR);
5617 if (programType != fileServer && !VCanUseFSSYNC()) {
5618 VChangeState_r(vp, VOL_STATE_ERROR);
5623 if (!vp->salvage.requested) {
5624 vp->salvage.requested = 1;
5625 vp->salvage.reason = reason;
5626 vp->stats.last_salvage = FT_ApproxTime();
5628 /* Note that it is not possible for us to reach this point if a
5629 * salvage is already running on this volume (even if the fileserver
5630 * was restarted during the salvage). If a salvage were running, the
5631 * salvager would have write-locked the volume header file, so when
5632 * we tried to lock the volume header, the lock would have failed,
5633 * and we would have failed during attachment prior to calling
5634 * VRequestSalvage. So we know that we can schedule salvages without
5635 * fear of a salvage already running for this volume. */
5637 if (vp->stats.salvages < SALVAGE_COUNT_MAX) {
5639 /* if we don't need to offline the volume, we can go directly
5640 * to SALVAGING. SALVAGING says the volume is offline and is
5641 * either salvaging or ready to be handed to the salvager.
5642 * SALVAGE_REQ says that we want to salvage the volume, but we
5643 * are waiting for it to go offline first. */
5644 if (flags & VOL_SALVAGE_NO_OFFLINE) {
5645 VChangeState_r(vp, VOL_STATE_SALVAGING);
5647 VChangeState_r(vp, VOL_STATE_SALVAGE_REQ);
5648 if (vp->nUsers == 0) {
5649 /* normally VOfflineForSalvage_r would be called from
5650 * PutVolume et al when nUsers reaches 0, but if
5651 * it's already 0, just do it ourselves, since PutVolume
5652 * isn't going to get called */
5653 VOfflineForSalvage_r(vp);
5656 /* If we are non-fileserver, we're telling the fileserver to
5657 * salvage the vol, so we don't need to give it back separately. */
5658 vp->needsPutBack = 0;
5662 Log("VRequestSalvage: volume %" AFS_VOLID_FMT " online salvaged too many times; forced offline.\n", afs_printable_VolumeId_lu(vp->hashid));
5664 /* make sure neither VScheduleSalvage_r nor
5665 * VUpdateSalvagePriority_r try to schedule another salvage */
5666 vp->salvage.requested = vp->salvage.scheduled = 0;
5668 VChangeState_r(vp, VOL_STATE_ERROR);
5677 * update salvageserver scheduling priority for a volume.
5679 * @param[in] vp pointer to volume object
5681 * @return operation status
5683 * @retval 1 request denied, or SALVSYNC communications failure
5685 * @pre VOL_LOCK is held.
5687 * @post in-core salvage priority counter is incremented. if at least
5688 * SALVAGE_PRIO_UPDATE_INTERVAL seconds have elapsed since the
5689 * last SALVSYNC_RAISEPRIO request, we contact the salvageserver
5690 * to update its priority queue. if no salvage is scheduled,
5691 * this function is a no-op.
5693 * @note DAFS fileserver only
5695 * @note this should be called whenever a VGetVolume fails due to a
5696 * pending salvage request
5698 * @todo should set exclusive state and drop glock around salvsync call
5700 * @internal volume package internal use only.
5703 VUpdateSalvagePriority_r(Volume * vp)
5707 #ifdef SALVSYNC_BUILD_CLIENT
5712 now = FT_ApproxTime();
5714 /* update the salvageserver priority queue occasionally so that
5715 * frequently requested volumes get moved to the head of the queue
5717 if ((vp->salvage.scheduled) &&
5718 (vp->stats.last_salvage_req < (now-SALVAGE_PRIO_UPDATE_INTERVAL))) {
5719 code = SALVSYNC_SalvageVolume(vp->hashid,
5720 VPartitionPath(vp->partition),
5725 vp->stats.last_salvage_req = now;
5726 if (code != SYNC_OK) {
5730 #endif /* SALVSYNC_BUILD_CLIENT */
5735 #if defined(SALVSYNC_BUILD_CLIENT) || defined(FSSYNC_BUILD_CLIENT)
5737 /* A couple of little helper functions. These return true if we tried to
5738 * use this mechanism to schedule a salvage, false if we haven't tried.
5739 * If we did try a salvage then the results are contained in code.
5743 try_SALVSYNC(Volume *vp, char *partName, int *code) {
5744 #ifdef SALVSYNC_BUILD_CLIENT
5745 if (VCanUseSALVSYNC()) {
5746 Log("Scheduling salvage for volume %" AFS_VOLID_FMT " on part %s over SALVSYNC\n",
5747 afs_printable_VolumeId_lu(vp->hashid), partName);
5749 /* can't use V_id() since there's no guarantee
5750 * we have the disk data header at this point */
5751 *code = SALVSYNC_SalvageVolume(vp->hashid,
5764 try_FSSYNC(Volume *vp, char *partName, int *code) {
5765 #ifdef FSSYNC_BUILD_CLIENT
5766 if (VCanUseFSSYNC()) {
5767 Log("Scheduling salvage for volume %" AFS_VOLID_FMT " on part %s over FSSYNC\n",
5768 afs_printable_VolumeId_lu(vp->hashid), partName);
5771 * If we aren't the fileserver, tell the fileserver the volume
5772 * needs to be salvaged. We could directly tell the
5773 * salvageserver, but the fileserver keeps track of some stats
5774 * related to salvages, and handles some other salvage-related
5775 * complications for us.
5777 *code = FSYNC_VolOp(vp->hashid, partName,
5778 FSYNC_VOL_FORCE_ERROR, FSYNC_SALVAGE, NULL);
5781 #endif /* FSSYNC_BUILD_CLIENT */
5786 * schedule a salvage with the salvage server or fileserver.
5788 * @param[in] vp pointer to volume object
5790 * @return operation status
5791 * @retval VCHECK_SALVAGE_OK (0) no pending salvage
5792 * @retval VCHECK_SALVAGE_SCHEDULED (1) salvage has been scheduled
5793 * @retval VCHECK_SALVAGE_ASYNC (2) salvage being scheduled
5794 * @retval VCHECK_SALVAGE_DENIED (3) salvage not scheduled; denied
5795 * @retval VCHECK_SALVAGE_FAIL (4) salvage not scheduled; failed
5798 * @arg VOL_LOCK is held.
5799 * @arg nUsers and nWaiters should be zero.
5801 * @post salvageserver or fileserver is sent a salvage request
5803 * @note If we are the fileserver, the request will be sent to the salvage
5804 * server over SALVSYNC. If we are not the fileserver, the request will be
5805 * sent to the fileserver over FSSYNC (FSYNC_VOL_FORCE_ERROR/FSYNC_SALVAGE).
5807 * @note the caller must check if the volume needs to be freed after calling
5808 * this; the volume may not have any references or be on any lists after
5809 * we return, and we do not free it
5813 * @internal volume package internal use only.
5816 VScheduleSalvage_r(Volume * vp)
5818 int ret = VCHECK_SALVAGE_SCHEDULED;
5820 VolState state_save;
5821 VThreadOptions_t * thread_opts;
5824 opr_Verify(VCanUseSALVSYNC() || VCanUseFSSYNC());
5826 if (vp->nWaiters || vp->nUsers) {
5827 return VCHECK_SALVAGE_ASYNC;
5830 /* prevent endless salvage,attach,salvage,attach,... loops */
5831 if (vp->stats.salvages >= SALVAGE_COUNT_MAX) {
5832 return VCHECK_SALVAGE_FAIL;
5836 * don't perform salvsync ops on certain threads
5838 thread_opts = pthread_getspecific(VThread_key);
5839 if (thread_opts == NULL) {
5840 thread_opts = &VThread_defaults;
5842 if (thread_opts->disallow_salvsync || vol_disallow_salvsync) {
5843 return VCHECK_SALVAGE_ASYNC;
5846 if (vp->salvage.scheduled) {
5847 return VCHECK_SALVAGE_SCHEDULED;
5850 VCreateReservation_r(vp);
5851 VWaitExclusiveState_r(vp);
5854 * XXX the scheduling process should really be done asynchronously
5855 * to avoid fssync deadlocks
5857 if (vp->salvage.scheduled) {
5858 ret = VCHECK_SALVAGE_SCHEDULED;
5860 /* if we haven't previously scheduled a salvage, do so now
5862 * set the volume to an exclusive state and drop the lock
5863 * around the SALVSYNC call
5865 strlcpy(partName, vp->partition->name, sizeof(partName));
5866 state_save = VChangeState_r(vp, VOL_STATE_SALVSYNC_REQ);
5869 opr_Verify(try_SALVSYNC(vp, partName, &code)
5870 || try_FSSYNC(vp, partName, &code));
5873 VChangeState_r(vp, state_save);
5875 if (code == SYNC_OK) {
5876 ret = VCHECK_SALVAGE_SCHEDULED;
5877 vp->salvage.scheduled = 1;
5878 vp->stats.last_salvage_req = FT_ApproxTime();
5879 if (VCanUseSALVSYNC()) {
5880 /* don't record these stats for non-fileservers; let the
5881 * fileserver take care of these */
5882 vp->stats.salvages++;
5883 IncUInt64(&VStats.salvages);
5887 case SYNC_BAD_COMMAND:
5888 case SYNC_COM_ERROR:
5889 ret = VCHECK_SALVAGE_FAIL;
5892 ret = VCHECK_SALVAGE_DENIED;
5893 Log("VScheduleSalvage_r: Salvage request for volume %" AFS_VOLID_FMT " "
5894 "denied\n", afs_printable_VolumeId_lu(vp->hashid));
5897 ret = VCHECK_SALVAGE_FAIL;
5898 Log("VScheduleSalvage_r: Salvage request for volume %" AFS_VOLID_FMT " "
5899 "failed\n", afs_printable_VolumeId_lu(vp->hashid));
5902 ret = VCHECK_SALVAGE_FAIL;
5903 Log("VScheduleSalvage_r: Salvage request for volume %" AFS_VOLID_FMT " "
5904 "received unknown protocol error %d\n",
5905 afs_printable_VolumeId_lu(vp->hashid), code);
5909 if (VCanUseFSSYNC()) {
5910 VChangeState_r(vp, VOL_STATE_ERROR);
5915 /* NB: this is cancelling the reservation we obtained above, but we do
5916 * not call VCancelReservation_r, since that may trigger the vp dtor,
5917 * possibly free'ing the vp. We need to keep the vp around after
5918 * this, as the caller may reference vp without any refs. Instead, it
5919 * is the duty of the caller to inspect 'vp' after we return to see if
5920 * needs to be freed. */
5921 opr_Verify(--vp->nWaiters >= 0);
5924 #endif /* SALVSYNC_BUILD_CLIENT || FSSYNC_BUILD_CLIENT */
5926 #ifdef SALVSYNC_BUILD_CLIENT
5929 * connect to the salvageserver SYNC service.
5931 * @return operation status
5935 * @post connection to salvageserver SYNC service established
5937 * @see VConnectSALV_r
5938 * @see VDisconnectSALV
5939 * @see VReconnectSALV
5946 retVal = VConnectSALV_r();
5952 * connect to the salvageserver SYNC service.
5954 * @return operation status
5958 * @pre VOL_LOCK is held.
5960 * @post connection to salvageserver SYNC service established
5963 * @see VDisconnectSALV_r
5964 * @see VReconnectSALV_r
5965 * @see SALVSYNC_clientInit
5967 * @internal volume package internal use only.
5970 VConnectSALV_r(void)
5972 return SALVSYNC_clientInit();
5976 * disconnect from the salvageserver SYNC service.
5978 * @return operation status
5981 * @pre client should have a live connection to the salvageserver
5983 * @post connection to salvageserver SYNC service destroyed
5985 * @see VDisconnectSALV_r
5987 * @see VReconnectSALV
5990 VDisconnectSALV(void)
5993 VDisconnectSALV_r();
5999 * disconnect from the salvageserver SYNC service.
6001 * @return operation status
6005 * @arg VOL_LOCK is held.
6006 * @arg client should have a live connection to the salvageserver.
6008 * @post connection to salvageserver SYNC service destroyed
6010 * @see VDisconnectSALV
6011 * @see VConnectSALV_r
6012 * @see VReconnectSALV_r
6013 * @see SALVSYNC_clientFinis
6015 * @internal volume package internal use only.
6018 VDisconnectSALV_r(void)
6020 return SALVSYNC_clientFinis();
6024 * disconnect and then re-connect to the salvageserver SYNC service.
6026 * @return operation status
6030 * @pre client should have a live connection to the salvageserver
6032 * @post old connection is dropped, and a new one is established
6035 * @see VDisconnectSALV
6036 * @see VReconnectSALV_r
6039 VReconnectSALV(void)
6043 retVal = VReconnectSALV_r();
6049 * disconnect and then re-connect to the salvageserver SYNC service.
6051 * @return operation status
6056 * @arg VOL_LOCK is held.
6057 * @arg client should have a live connection to the salvageserver.
6059 * @post old connection is dropped, and a new one is established
6061 * @see VConnectSALV_r
6062 * @see VDisconnectSALV
6063 * @see VReconnectSALV
6064 * @see SALVSYNC_clientReconnect
6066 * @internal volume package internal use only.
6069 VReconnectSALV_r(void)
6071 return SALVSYNC_clientReconnect();
6073 #endif /* SALVSYNC_BUILD_CLIENT */
6074 #endif /* AFS_DEMAND_ATTACH_FS */
6077 /***************************************************/
6078 /* FSSYNC routines */
6079 /***************************************************/
6081 /* This must be called by any volume utility which needs to run while the
6082 file server is also running. This is separated from VInitVolumePackage2 so
6083 that a utility can fork--and each of the children can independently
6084 initialize communication with the file server */
6085 #ifdef FSSYNC_BUILD_CLIENT
6087 * connect to the fileserver SYNC service.
6089 * @return operation status
6094 * @arg VInit must equal 2.
6095 * @arg Program Type must not be fileserver or salvager.
6097 * @post connection to fileserver SYNC service established
6100 * @see VDisconnectFS
6101 * @see VChildProcReconnectFS
6108 retVal = VConnectFS_r();
6114 * connect to the fileserver SYNC service.
6116 * @return operation status
6121 * @arg VInit must equal 2.
6122 * @arg Program Type must not be fileserver or salvager.
6123 * @arg VOL_LOCK is held.
6125 * @post connection to fileserver SYNC service established
6128 * @see VDisconnectFS_r
6129 * @see VChildProcReconnectFS_r
6131 * @internal volume package internal use only.
6137 opr_Assert((VInit == 2) &&
6138 (programType != fileServer) &&
6139 (programType != salvager));
6140 rc = FSYNC_clientInit();
6148 * disconnect from the fileserver SYNC service.
6151 * @arg client should have a live connection to the fileserver.
6152 * @arg VOL_LOCK is held.
6153 * @arg Program Type must not be fileserver or salvager.
6155 * @post connection to fileserver SYNC service destroyed
6157 * @see VDisconnectFS
6159 * @see VChildProcReconnectFS_r
6161 * @internal volume package internal use only.
6164 VDisconnectFS_r(void)
6166 opr_Assert((programType != fileServer) &&
6167 (programType != salvager));
6168 FSYNC_clientFinis();
6173 * disconnect from the fileserver SYNC service.
6176 * @arg client should have a live connection to the fileserver.
6177 * @arg Program Type must not be fileserver or salvager.
6179 * @post connection to fileserver SYNC service destroyed
6181 * @see VDisconnectFS_r
6183 * @see VChildProcReconnectFS
6194 * connect to the fileserver SYNC service from a child process following a fork.
6196 * @return operation status
6201 * @arg VOL_LOCK is held.
6202 * @arg current FSYNC handle is shared with a parent process
6204 * @post current FSYNC handle is discarded and a new connection to the
6205 * fileserver SYNC service is established
6207 * @see VChildProcReconnectFS
6209 * @see VDisconnectFS_r
6211 * @internal volume package internal use only.
6214 VChildProcReconnectFS_r(void)
6216 return FSYNC_clientChildProcReconnect();
6220 * connect to the fileserver SYNC service from a child process following a fork.
6222 * @return operation status
6226 * @pre current FSYNC handle is shared with a parent process
6228 * @post current FSYNC handle is discarded and a new connection to the
6229 * fileserver SYNC service is established
6231 * @see VChildProcReconnectFS_r
6233 * @see VDisconnectFS
6236 VChildProcReconnectFS(void)
6240 ret = VChildProcReconnectFS_r();
6244 #endif /* FSSYNC_BUILD_CLIENT */
6247 /***************************************************/
6248 /* volume bitmap routines */
6249 /***************************************************/
6252 * Grow the bitmap by the defined increment
6255 VGrowBitmap(struct vnodeIndex *index)
6259 bp = realloc(index->bitmap, index->bitmapSize + VOLUME_BITMAP_GROWSIZE);
6260 osi_Assert(bp != NULL);
6262 bp += index->bitmapSize;
6263 memset(bp, 0, VOLUME_BITMAP_GROWSIZE);
6264 index->bitmapOffset = index->bitmapSize;
6265 index->bitmapSize += VOLUME_BITMAP_GROWSIZE;
6271 * allocate a vnode bitmap number for the vnode
6273 * @param[out] ec error code
6274 * @param[in] vp volume object pointer
6275 * @param[in] index vnode index number for the vnode
6276 * @param[in] flags flag values described in note
6278 * @note for DAFS, flags parameter controls locking behavior.
6279 * If (flags & VOL_ALLOC_BITMAP_WAIT) is set, then this function
6280 * will create a reservation and block on any other exclusive
6281 * operations. Otherwise, this function assumes the caller
6282 * already has exclusive access to vp, and we just change the
6285 * @pre VOL_LOCK held
6287 * @return bit number allocated
6293 VAllocBitmapEntry_r(Error * ec, Volume * vp,
6294 struct vnodeIndex *index, int flags)
6298 #ifdef AFS_DEMAND_ATTACH_FS
6299 VolState state_save;
6300 #endif /* AFS_DEMAND_ATTACH_FS */
6304 /* This test is probably redundant */
6305 if (!VolumeWriteable(vp)) {
6306 *ec = (bit32) VREADONLY;
6310 #ifdef AFS_DEMAND_ATTACH_FS
6311 if (flags & VOL_ALLOC_BITMAP_WAIT) {
6312 VCreateReservation_r(vp);
6313 VWaitExclusiveState_r(vp);
6315 state_save = VChangeState_r(vp, VOL_STATE_GET_BITMAP);
6316 #endif /* AFS_DEMAND_ATTACH_FS */
6319 if ((programType == fileServer) && !index->bitmap) {
6321 #ifndef AFS_DEMAND_ATTACH_FS
6322 /* demand attach fs uses the volume state to avoid races.
6323 * specialStatus field is not used at all */
6325 if (vp->specialStatus == VBUSY) {
6326 if (vp->goingOffline) { /* vos dump waiting for the volume to
6327 * go offline. We probably come here
6328 * from AddNewReadableResidency */
6331 while (vp->specialStatus == VBUSY) {
6332 #ifdef AFS_PTHREAD_ENV
6336 #else /* !AFS_PTHREAD_ENV */
6338 #endif /* !AFS_PTHREAD_ENV */
6342 #endif /* !AFS_DEMAND_ATTACH_FS */
6344 if (!index->bitmap) {
6345 #ifndef AFS_DEMAND_ATTACH_FS
6346 vp->specialStatus = VBUSY; /* Stop anyone else from using it. */
6347 #endif /* AFS_DEMAND_ATTACH_FS */
6348 for (i = 0; i < nVNODECLASSES; i++) {
6349 VGetBitmap_r(ec, vp, i);
6351 #ifdef AFS_DEMAND_ATTACH_FS
6352 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, 0 /*flags*/);
6353 #else /* AFS_DEMAND_ATTACH_FS */
6354 DeleteVolumeFromHashTable(vp);
6355 vp->shuttingDown = 1; /* Let who has it free it. */
6356 vp->specialStatus = 0;
6357 #endif /* AFS_DEMAND_ATTACH_FS */
6361 #ifndef AFS_DEMAND_ATTACH_FS
6363 vp->specialStatus = 0; /* Allow others to have access. */
6364 #endif /* AFS_DEMAND_ATTACH_FS */
6367 #endif /* BITMAP_LATER */
6369 #ifdef AFS_DEMAND_ATTACH_FS
6371 #endif /* AFS_DEMAND_ATTACH_FS */
6372 bp = index->bitmap + index->bitmapOffset;
6373 ep = index->bitmap + index->bitmapSize;
6375 if ((*(bit32 *) bp) != (bit32) 0xffffffff) {
6377 index->bitmapOffset = (afs_uint32) (bp - index->bitmap);
6380 o = ffs(~*bp) - 1; /* ffs is documented in BSTRING(3) */
6382 ret = ((bp - index->bitmap) * 8 + o);
6383 #ifdef AFS_DEMAND_ATTACH_FS
6385 #endif /* AFS_DEMAND_ATTACH_FS */
6388 bp += sizeof(bit32) /* i.e. 4 */ ;
6390 /* No bit map entry--must grow bitmap */
6392 bp = index->bitmap + index->bitmapOffset;
6394 ret = index->bitmapOffset * 8;
6395 #ifdef AFS_DEMAND_ATTACH_FS
6397 #endif /* AFS_DEMAND_ATTACH_FS */
6400 #ifdef AFS_DEMAND_ATTACH_FS
6401 VChangeState_r(vp, state_save);
6402 if (flags & VOL_ALLOC_BITMAP_WAIT) {
6403 VCancelReservation_r(vp);
6405 #endif /* AFS_DEMAND_ATTACH_FS */
6410 VAllocBitmapEntry(Error * ec, Volume * vp, struct vnodeIndex * index)
6414 retVal = VAllocBitmapEntry_r(ec, vp, index, VOL_ALLOC_BITMAP_WAIT);
6420 VFreeBitMapEntry_r(Error * ec, Volume *vp, struct vnodeIndex *index,
6421 unsigned bitNumber, int flags)
6423 unsigned int offset;
6427 #ifdef AFS_DEMAND_ATTACH_FS
6428 if (flags & VOL_FREE_BITMAP_WAIT) {
6429 /* VAllocBitmapEntry_r allocs bitmap entries under an exclusive volume
6430 * state, so ensure we're not in an exclusive volume state when we update
6432 VCreateReservation_r(vp);
6433 VWaitExclusiveState_r(vp);
6440 #endif /* BITMAP_LATER */
6442 offset = bitNumber >> 3;
6443 if (offset >= index->bitmapSize) {
6447 if (offset < index->bitmapOffset)
6448 index->bitmapOffset = offset & ~3; /* Truncate to nearest bit32 */
6449 *(index->bitmap + offset) &= ~(1 << (bitNumber & 0x7));
6452 #ifdef AFS_DEMAND_ATTACH_FS
6453 VCancelReservation_r(vp);
6455 return; /* make the compiler happy for non-DAFS */
6459 VFreeBitMapEntry(Error * ec, Volume *vp, struct vnodeIndex *index,
6463 VFreeBitMapEntry_r(ec, vp, index, bitNumber, VOL_FREE_BITMAP_WAIT);
6467 /* this function will drop the glock internally.
6468 * for old pthread fileservers, this is safe thanks to vbusy.
6470 * for demand attach fs, caller must have already called
6471 * VCreateReservation_r and VWaitExclusiveState_r */
6473 VGetBitmap_r(Error * ec, Volume * vp, VnodeClass class)
6475 StreamHandle_t *file;
6476 afs_sfsize_t nVnodes, size;
6477 struct VnodeClassInfo *vcp = &VnodeClassInfo[class];
6478 struct vnodeIndex *vip = &vp->vnodeIndex[class];
6479 struct VnodeDiskObject *vnode;
6480 unsigned int unique = 0;
6484 #endif /* BITMAP_LATER */
6485 #ifdef AFS_DEMAND_ATTACH_FS
6486 VolState state_save;
6487 #endif /* AFS_DEMAND_ATTACH_FS */
6491 #ifdef AFS_DEMAND_ATTACH_FS
6492 state_save = VChangeState_r(vp, VOL_STATE_GET_BITMAP);
6493 #endif /* AFS_DEMAND_ATTACH_FS */
6496 fdP = IH_OPEN(vip->handle);
6497 opr_Assert(fdP != NULL);
6498 file = FDH_FDOPEN(fdP, "r");
6499 opr_Assert(file != NULL);
6500 vnode = malloc(vcp->diskSize);
6501 opr_Assert(vnode != NULL);
6502 size = OS_SIZE(fdP->fd_fd);
6503 opr_Assert(size != -1);
6504 nVnodes = (size <= vcp->diskSize ? 0 : size - vcp->diskSize)
6506 vip->bitmapSize = ((nVnodes / 8) + 10) / 4 * 4; /* The 10 is a little extra so
6507 * a few files can be created in this volume,
6508 * the whole thing is rounded up to nearest 4
6509 * bytes, because the bit map allocator likes
6512 BitMap = (byte *) calloc(1, vip->bitmapSize);
6513 opr_Assert(BitMap != NULL);
6514 #else /* BITMAP_LATER */
6515 vip->bitmap = (byte *) calloc(1, vip->bitmapSize);
6516 opr_Assert(vip->bitmap != NULL);
6517 vip->bitmapOffset = 0;
6518 #endif /* BITMAP_LATER */
6519 if (STREAM_ASEEK(file, vcp->diskSize) != -1) {
6521 for (bitNumber = 0; bitNumber < nVnodes + 100; bitNumber++) {
6522 if (STREAM_READ(vnode, vcp->diskSize, 1, file) != 1)
6524 if (vnode->type != vNull) {
6525 if (vnode->vnodeMagic != vcp->magic) {
6526 Log("GetBitmap: addled vnode index in volume %s; volume needs salvage\n", V_name(vp));
6531 *(BitMap + (bitNumber >> 3)) |= (1 << (bitNumber & 0x7));
6532 #else /* BITMAP_LATER */
6533 *(vip->bitmap + (bitNumber >> 3)) |= (1 << (bitNumber & 0x7));
6534 #endif /* BITMAP_LATER */
6535 if (unique <= vnode->uniquifier)
6536 unique = vnode->uniquifier + 1;
6538 #ifndef AFS_PTHREAD_ENV
6539 if ((bitNumber & 0x00ff) == 0x0ff) { /* every 256 iterations */
6542 #endif /* !AFS_PTHREAD_ENV */
6545 if (vp->nextVnodeUnique < unique) {
6546 Log("GetBitmap: bad volume uniquifier for volume %s; volume needs salvage\n", V_name(vp));
6549 /* Paranoia, partly justified--I think fclose after fdopen
6550 * doesn't seem to close fd. In any event, the documentation
6551 * doesn't specify, so it's safer to close it twice.
6559 /* There may have been a racing condition with some other thread, both
6560 * creating the bitmaps for this volume. If the other thread was faster
6561 * the pointer to bitmap should already be filled and we can free ours.
6563 if (vip->bitmap == NULL) {
6564 vip->bitmap = BitMap;
6565 vip->bitmapOffset = 0;
6568 #endif /* BITMAP_LATER */
6569 #ifdef AFS_DEMAND_ATTACH_FS
6570 VChangeState_r(vp, state_save);
6571 #endif /* AFS_DEMAND_ATTACH_FS */
6575 /***************************************************/
6576 /* Volume Path and Volume Number utility routines */
6577 /***************************************************/
6580 * find the first occurrence of a volume header file and return the path.
6582 * @param[out] ec outbound error code
6583 * @param[in] volumeId volume id to find
6584 * @param[out] partitionp pointer to disk partition path string
6585 * @param[out] namep pointer to volume header file name string
6587 * @post path to first occurrence of volume header is returned in partitionp
6588 * and namep, or ec is set accordingly.
6590 * @warning this function is NOT re-entrant -- partitionp and namep point to
6591 * static data segments
6593 * @note if a volume utility inadvertently leaves behind a stale volume header
6594 * on a vice partition, it is possible for callers to get the wrong one,
6595 * depending on the order of the disk partition linked list.
6599 VGetVolumePath(Error * ec, VolumeId volumeId, char **partitionp, char **namep)
6601 static char partition[VMAXPATHLEN], name[VMAXPATHLEN];
6602 char path[VMAXPATHLEN];
6604 struct DiskPartition64 *dp;
6607 name[0] = OS_DIRSEPC;
6608 snprintf(&name[1], (sizeof name) - 1, VFORMAT,
6609 afs_printable_VolumeId_lu(volumeId));
6610 for (dp = DiskPartitionList; dp; dp = dp->next) {
6611 struct afs_stat_st status;
6612 strcpy(path, VPartitionPath(dp));
6614 if (afs_stat(path, &status) == 0) {
6615 strcpy(partition, dp->name);
6622 *partitionp = *namep = NULL;
6624 *partitionp = partition;
6630 * extract a volume number from a volume header filename string.
6632 * @param[in] name volume header filename string
6634 * @return volume number
6636 * @note the string must be of the form VFORMAT. the only permissible
6637 * deviation is a leading OS_DIRSEPC character.
6642 VolumeNumber(char *name)
6644 if (*name == OS_DIRSEPC)
6646 return strtoul(name + 1, NULL, 10);
6650 * compute the volume header filename.
6652 * @param[in] volumeId
6654 * @return volume header filename
6656 * @post volume header filename string is constructed
6658 * @warning this function is NOT re-entrant -- the returned string is
6659 * stored in a static char array. see VolumeExternalName_r
6660 * for a re-entrant equivalent.
6662 * @see VolumeExternalName_r
6664 * @deprecated due to the above re-entrancy warning, this interface should
6665 * be considered deprecated. Please use VolumeExternalName_r
6669 VolumeExternalName(VolumeId volumeId)
6671 static char name[VMAXPATHLEN];
6672 snprintf(name, sizeof name, VFORMAT, afs_printable_VolumeId_lu(volumeId));
6677 * compute the volume header filename.
6679 * @param[in] volumeId
6680 * @param[inout] name array in which to store filename
6681 * @param[in] len length of name array
6683 * @return result code from afs_snprintf
6685 * @see VolumeExternalName
6688 * @note re-entrant equivalent of VolumeExternalName
6691 VolumeExternalName_r(VolumeId volumeId, char * name, size_t len)
6693 return snprintf(name, len, VFORMAT, afs_printable_VolumeId_lu(volumeId));
6697 /***************************************************/
6698 /* Volume Usage Statistics routines */
6699 /***************************************************/
6701 #define OneDay (86400) /* 24 hours' worth of seconds */
6704 Midnight(time_t t) {
6705 struct tm local, *l;
6708 #if defined(AFS_PTHREAD_ENV) && !defined(AFS_NT40_ENV)
6709 l = localtime_r(&t, &local);
6715 /* the following is strictly speaking problematic on the
6716 switching day to daylight saving time, after the switch,
6717 as tm_isdst does not match. Similarly, on the looong day when
6718 switching back the OneDay check will not do what naively expected!
6719 The effects are minor, though, and more a matter of interpreting
6721 #ifndef AFS_PTHREAD_ENV
6724 local.tm_hour = local.tm_min=local.tm_sec = 0;
6725 midnight = mktime(&local);
6726 if (midnight != (time_t) -1) return(midnight);
6728 return( (t/OneDay)*OneDay );
6732 /*------------------------------------------------------------------------
6733 * [export] VAdjustVolumeStatistics
6736 * If we've passed midnight, we need to update all the day use
6737 * statistics as well as zeroing the detailed volume statistics
6738 * (if we are implementing them).
6741 * vp : Pointer to the volume structure describing the lucky
6742 * volume being considered for update.
6748 * Nothing interesting.
6752 *------------------------------------------------------------------------*/
6755 VAdjustVolumeStatistics_r(Volume * vp)
6757 unsigned int now = FT_ApproxTime();
6759 if (now - V_dayUseDate(vp) > OneDay) {
6762 ndays = (now - V_dayUseDate(vp)) / OneDay;
6763 for (i = 6; i > ndays - 1; i--)
6764 V_weekUse(vp)[i] = V_weekUse(vp)[i - ndays];
6765 for (i = 0; i < ndays - 1 && i < 7; i++)
6766 V_weekUse(vp)[i] = 0;
6768 V_weekUse(vp)[ndays - 1] = V_dayUse(vp);
6770 V_dayUseDate(vp) = Midnight(now);
6773 * All we need to do is bzero the entire VOL_STATS_BYTES of
6774 * the detailed volume statistics area.
6776 memset((V_stat_area(vp)), 0, VOL_STATS_BYTES);
6779 /*It's been more than a day of collection */
6781 * Always return happily.
6784 } /*VAdjustVolumeStatistics */
6787 VAdjustVolumeStatistics(Volume * vp)
6791 retVal = VAdjustVolumeStatistics_r(vp);
6797 VBumpVolumeUsage_r(Volume * vp)
6799 unsigned int now = FT_ApproxTime();
6800 V_accessDate(vp) = now;
6801 if (now - V_dayUseDate(vp) > OneDay)
6802 VAdjustVolumeStatistics_r(vp);
6804 * Save the volume header image to disk after a threshold of bumps to dayUse,
6805 * at most every usage_rate_limit seconds.
6808 vp->usage_bumps_outstanding++;
6809 if (vp->usage_bumps_outstanding >= vol_opts.usage_threshold
6810 && vp->usage_bumps_next_write <= now) {
6812 vp->usage_bumps_outstanding = 0;
6813 vp->usage_bumps_next_write = now + vol_opts.usage_rate_limit;
6814 VUpdateVolume_r(&error, vp, VOL_UPDATE_WAIT);
6819 VBumpVolumeUsage(Volume * vp)
6822 VBumpVolumeUsage_r(vp);
6827 VSetDiskUsage_r(void)
6829 #ifndef AFS_DEMAND_ATTACH_FS
6830 static int FifteenMinuteCounter = 0;
6834 /* NOTE: Don't attempt to access the partitions list until the
6835 * initialization level indicates that all volumes are attached,
6836 * which implies that all partitions are initialized. */
6837 #ifdef AFS_PTHREAD_ENV
6838 VOL_CV_WAIT(&vol_vinit_cond);
6839 #else /* AFS_PTHREAD_ENV */
6841 #endif /* AFS_PTHREAD_ENV */
6844 VResetDiskUsage_r();
6846 #ifndef AFS_DEMAND_ATTACH_FS
6847 if (++FifteenMinuteCounter == 3) {
6848 FifteenMinuteCounter = 0;
6851 #endif /* !AFS_DEMAND_ATTACH_FS */
6863 /***************************************************/
6864 /* Volume Update List routines */
6865 /***************************************************/
6867 /* The number of minutes that a volume hasn't been updated before the
6868 * "Dont salvage" flag in the volume header will be turned on */
6869 #define SALVAGE_INTERVAL (10*60)
6874 * volume update list functionality has been moved into the VLRU
6875 * the DONT_SALVAGE flag is now set during VLRU demotion
6878 #ifndef AFS_DEMAND_ATTACH_FS
6879 static VolumeId *UpdateList = NULL; /* Pointer to array of Volume ID's */
6880 static int nUpdatedVolumes = 0; /* Updated with entry in UpdateList, salvage after crash flag on */
6881 static int updateSize = 0; /* number of entries possible */
6882 #define UPDATE_LIST_SIZE 128 /* initial size increment (must be a power of 2!) */
6883 #endif /* !AFS_DEMAND_ATTACH_FS */
6886 VAddToVolumeUpdateList_r(Error * ec, Volume * vp)
6889 vp->updateTime = FT_ApproxTime();
6890 if (V_dontSalvage(vp) == 0)
6892 V_dontSalvage(vp) = 0;
6893 VSyncVolume_r(ec, vp, 0);
6894 #ifdef AFS_DEMAND_ATTACH_FS
6895 V_attachFlags(vp) &= ~(VOL_HDR_DONTSALV);
6896 #else /* !AFS_DEMAND_ATTACH_FS */
6899 if (UpdateList == NULL) {
6900 updateSize = UPDATE_LIST_SIZE;
6901 UpdateList = malloc(sizeof(VolumeId) * updateSize);
6903 if (nUpdatedVolumes == updateSize) {
6905 if (updateSize > 524288) {
6906 Log("warning: there is likely a bug in the volume update scanner\n");
6909 UpdateList = realloc(UpdateList,
6910 sizeof(VolumeId) * updateSize);
6913 opr_Assert(UpdateList != NULL);
6914 UpdateList[nUpdatedVolumes++] = V_id(vp);
6915 #endif /* !AFS_DEMAND_ATTACH_FS */
6918 #ifndef AFS_DEMAND_ATTACH_FS
6920 VScanUpdateList(void)
6925 afs_uint32 now = FT_ApproxTime();
6926 /* Be careful with this code, since it works with interleaved calls to AddToVolumeUpdateList */
6927 for (i = gap = 0; i < nUpdatedVolumes; i++) {
6929 UpdateList[i - gap] = UpdateList[i];
6931 /* XXX this routine needlessly messes up the Volume LRU by
6932 * breaking the LRU temporal-locality assumptions.....
6933 * we should use a special volume header allocator here */
6934 vp = VGetVolume_r(&error, UpdateList[i - gap] = UpdateList[i]);
6937 } else if (vp->nUsers == 1 && now - vp->updateTime > SALVAGE_INTERVAL) {
6938 V_dontSalvage(vp) = DONT_SALVAGE;
6939 VUpdateVolume_r(&error, vp, 0); /* No need to fsync--not critical */
6947 #ifndef AFS_PTHREAD_ENV
6949 #endif /* !AFS_PTHREAD_ENV */
6951 nUpdatedVolumes -= gap;
6953 #endif /* !AFS_DEMAND_ATTACH_FS */
6956 /***************************************************/
6957 /* Volume LRU routines */
6958 /***************************************************/
6963 * with demand attach fs, we attempt to soft detach(1)
6964 * volumes which have not been accessed in a long time
6965 * in order to speed up fileserver shutdown
6967 * (1) by soft detach we mean a process very similar
6968 * to VOffline, except the final state of the
6969 * Volume will be VOL_STATE_PREATTACHED, instead
6970 * of the usual VOL_STATE_UNATTACHED
6972 #ifdef AFS_DEMAND_ATTACH_FS
6974 /* implementation is reminiscent of a generational GC
6976 * queue 0 is newly attached volumes. this queue is
6977 * sorted by attach timestamp
6979 * queue 1 is volumes that have been around a bit
6980 * longer than queue 0. this queue is sorted by
6983 * queue 2 is volumes tha have been around the longest.
6984 * this queue is unsorted
6986 * queue 3 is volumes that have been marked as
6987 * candidates for soft detachment. this queue is
6990 #define VLRU_GENERATIONS 3 /**< number of generations in VLRU */
6991 #define VLRU_QUEUES 5 /**< total number of VLRU queues */
6994 * definition of a VLRU queue.
6997 volatile struct rx_queue q;
7004 * main VLRU data structure.
7007 struct VLRU_q q[VLRU_QUEUES]; /**< VLRU queues */
7010 /** time interval (in seconds) between promotion passes for
7011 * each young generation queue. */
7012 afs_uint32 promotion_interval[VLRU_GENERATIONS-1];
7014 /** time interval (in seconds) between soft detach candidate
7015 * scans for each generation queue.
7017 * scan_interval[VLRU_QUEUE_CANDIDATE] defines how frequently
7018 * we perform a soft detach pass. */
7019 afs_uint32 scan_interval[VLRU_GENERATIONS+1];
7021 /* scheduler state */
7022 int next_idx; /**< next queue to receive attention */
7023 afs_uint32 last_promotion[VLRU_GENERATIONS-1]; /**< timestamp of last promotion scan */
7024 afs_uint32 last_scan[VLRU_GENERATIONS+1]; /**< timestamp of last detach scan */
7026 int scanner_state; /**< state of scanner thread */
7027 pthread_cond_t cv; /**< state transition CV */
7030 /** global VLRU state */
7031 static struct VLRU volume_LRU;
7034 * defined states for VLRU scanner thread.
7037 VLRU_SCANNER_STATE_OFFLINE = 0, /**< vlru scanner thread is offline */
7038 VLRU_SCANNER_STATE_ONLINE = 1, /**< vlru scanner thread is online */
7039 VLRU_SCANNER_STATE_SHUTTING_DOWN = 2, /**< vlru scanner thread is shutting down */
7040 VLRU_SCANNER_STATE_PAUSING = 3, /**< vlru scanner thread is getting ready to pause */
7041 VLRU_SCANNER_STATE_PAUSED = 4 /**< vlru scanner thread is paused */
7042 } vlru_thread_state_t;
7044 /* vlru disk data header stuff */
7045 #define VLRU_DISK_MAGIC 0x7a8b9cad /**< vlru disk entry magic number */
7046 #define VLRU_DISK_VERSION 1 /**< vlru disk entry version number */
7048 /** vlru default expiration time (for eventual fs state serialization of vlru data) */
7049 #define VLRU_DUMP_EXPIRATION_TIME (60*60*24*7) /* expire vlru data after 1 week */
7052 /** minimum volume inactivity (in seconds) before a volume becomes eligible for
7053 * soft detachment. */
7054 static afs_uint32 VLRU_offline_thresh = VLRU_DEFAULT_OFFLINE_THRESH;
7056 /** time interval (in seconds) between VLRU scanner thread soft detach passes. */
7057 static afs_uint32 VLRU_offline_interval = VLRU_DEFAULT_OFFLINE_INTERVAL;
7059 /** maximum number of volumes to soft detach in a VLRU soft detach pass. */
7060 static afs_uint32 VLRU_offline_max = VLRU_DEFAULT_OFFLINE_MAX;
7062 /** VLRU control flag. non-zero value implies VLRU subsystem is activated. */
7063 static afs_uint32 VLRU_enabled = 1;
7065 /* queue synchronization routines */
7066 static void VLRU_BeginExclusive_r(struct VLRU_q * q);
7067 static void VLRU_EndExclusive_r(struct VLRU_q * q);
7068 static void VLRU_Wait_r(struct VLRU_q * q);
7071 * set VLRU subsystem tunable parameters.
7073 * @param[in] option tunable option to modify
7074 * @param[in] val new value for tunable parameter
7076 * @pre @c VInitVolumePackage2 has not yet been called.
7078 * @post tunable parameter is modified
7082 * @note valid option parameters are:
7083 * @arg @c VLRU_SET_THRESH
7084 * set the period of inactivity after which
7085 * volumes are eligible for soft detachment
7086 * @arg @c VLRU_SET_INTERVAL
7087 * set the time interval between calls
7088 * to the volume LRU "garbage collector"
7089 * @arg @c VLRU_SET_MAX
7090 * set the max number of volumes to deallocate
7094 VLRU_SetOptions(int option, afs_uint32 val)
7096 if (option == VLRU_SET_THRESH) {
7097 VLRU_offline_thresh = val;
7098 } else if (option == VLRU_SET_INTERVAL) {
7099 VLRU_offline_interval = val;
7100 } else if (option == VLRU_SET_MAX) {
7101 VLRU_offline_max = val;
7102 } else if (option == VLRU_SET_ENABLED) {
7105 VLRU_ComputeConstants();
7109 * compute VLRU internal timing parameters.
7111 * @post VLRU scanner thread internal timing parameters are computed
7113 * @note computes internal timing parameters based upon user-modifiable
7114 * tunable parameters.
7118 * @internal volume package internal use only.
7121 VLRU_ComputeConstants(void)
7123 afs_uint32 factor = VLRU_offline_thresh / VLRU_offline_interval;
7125 /* compute the candidate scan interval */
7126 volume_LRU.scan_interval[VLRU_QUEUE_CANDIDATE] = VLRU_offline_interval;
7128 /* compute the promotion intervals */
7129 volume_LRU.promotion_interval[VLRU_QUEUE_NEW] = VLRU_offline_thresh * 2;
7130 volume_LRU.promotion_interval[VLRU_QUEUE_MID] = VLRU_offline_thresh * 4;
7133 /* compute the gen 0 scan interval */
7134 volume_LRU.scan_interval[VLRU_QUEUE_NEW] = VLRU_offline_thresh / 8;
7136 /* compute the gen 0 scan interval */
7137 volume_LRU.scan_interval[VLRU_QUEUE_NEW] = VLRU_offline_interval * 2;
7142 * initialize VLRU subsystem.
7144 * @pre this function has not yet been called
7146 * @post VLRU subsystem is initialized and VLRU scanner thread is starting
7150 * @internal volume package internal use only.
7156 pthread_attr_t attrs;
7159 if (!VLRU_enabled) {
7160 Log("VLRU: disabled\n");
7164 /* initialize each of the VLRU queues */
7165 for (i = 0; i < VLRU_QUEUES; i++) {
7166 queue_Init(&volume_LRU.q[i]);
7167 volume_LRU.q[i].len = 0;
7168 volume_LRU.q[i].busy = 0;
7169 opr_cv_init(&volume_LRU.q[i].cv);
7172 /* setup the timing constants */
7173 VLRU_ComputeConstants();
7175 /* XXX put inside LogLevel check? */
7176 Log("VLRU: starting scanner with the following configuration parameters:\n");
7177 Log("VLRU: offlining volumes after minimum of %d seconds of inactivity\n", VLRU_offline_thresh);
7178 Log("VLRU: running VLRU soft detach pass every %d seconds\n", VLRU_offline_interval);
7179 Log("VLRU: taking up to %d volumes offline per pass\n", VLRU_offline_max);
7180 Log("VLRU: scanning generation 0 for inactive volumes every %d seconds\n", volume_LRU.scan_interval[0]);
7181 Log("VLRU: scanning for promotion/demotion between generations 0 and 1 every %d seconds\n", volume_LRU.promotion_interval[0]);
7182 Log("VLRU: scanning for promotion/demotion between generations 1 and 2 every %d seconds\n", volume_LRU.promotion_interval[1]);
7184 /* start up the VLRU scanner */
7185 volume_LRU.scanner_state = VLRU_SCANNER_STATE_OFFLINE;
7186 if (programType == fileServer) {
7187 opr_cv_init(&volume_LRU.cv);
7188 opr_Verify(pthread_attr_init(&attrs) == 0);
7189 opr_Verify(pthread_attr_setdetachstate(&attrs,
7190 PTHREAD_CREATE_DETACHED) == 0);
7191 opr_Verify(pthread_create(&tid, &attrs,
7192 &VLRU_ScannerThread, NULL) == 0);
7197 * initialize the VLRU-related fields of a newly allocated volume object.
7199 * @param[in] vp pointer to volume object
7202 * @arg @c VOL_LOCK is held.
7203 * @arg volume object is not on a VLRU queue.
7205 * @post VLRU fields are initialized to indicate that volume object is not
7206 * currently registered with the VLRU subsystem
7210 * @internal volume package interal use only.
7213 VLRU_Init_Node_r(Volume * vp)
7218 opr_Assert(queue_IsNotOnQueue(&vp->vlru));
7219 vp->vlru.idx = VLRU_QUEUE_INVALID;
7223 * add a volume object to a VLRU queue.
7225 * @param[in] vp pointer to volume object
7228 * @arg @c VOL_LOCK is held.
7229 * @arg caller MUST hold a lightweight ref on @p vp.
7230 * @arg caller MUST NOT hold exclusive ownership of the VLRU queue.
7232 * @post the volume object is added to the appropriate VLRU queue
7234 * @note if @c vp->vlru.idx contains the index of a valid VLRU queue,
7235 * then the volume is added to that queue. Otherwise, the value
7236 * @c VLRU_QUEUE_NEW is stored into @c vp->vlru.idx and the
7237 * volume is added to the NEW generation queue.
7239 * @note @c VOL_LOCK may be dropped internally
7241 * @note Volume state is temporarily set to @c VOL_STATE_VLRU_ADD
7242 * during the add operation, and is restored to the previous
7243 * state prior to return.
7247 * @internal volume package internal use only.
7250 VLRU_Add_r(Volume * vp)
7253 VolState state_save;
7258 if (queue_IsOnQueue(&vp->vlru))
7261 state_save = VChangeState_r(vp, VOL_STATE_VLRU_ADD);
7264 if ((idx < 0) || (idx >= VLRU_QUEUE_INVALID)) {
7265 idx = VLRU_QUEUE_NEW;
7268 VLRU_Wait_r(&volume_LRU.q[idx]);
7270 /* repeat check since VLRU_Wait_r may have dropped
7272 if (queue_IsNotOnQueue(&vp->vlru)) {
7274 queue_Prepend(&volume_LRU.q[idx], &vp->vlru);
7275 volume_LRU.q[idx].len++;
7276 V_attachFlags(vp) |= VOL_ON_VLRU;
7277 vp->stats.last_promote = FT_ApproxTime();
7280 VChangeState_r(vp, state_save);
7284 * delete a volume object from a VLRU queue.
7286 * @param[in] vp pointer to volume object
7289 * @arg @c VOL_LOCK is held.
7290 * @arg caller MUST hold a lightweight ref on @p vp.
7291 * @arg caller MUST NOT hold exclusive ownership of the VLRU queue.
7293 * @post volume object is removed from the VLRU queue
7295 * @note @c VOL_LOCK may be dropped internally
7299 * @todo We should probably set volume state to something exlcusive
7300 * (as @c VLRU_Add_r does) prior to dropping @c VOL_LOCK.
7302 * @internal volume package internal use only.
7305 VLRU_Delete_r(Volume * vp)
7312 if (queue_IsNotOnQueue(&vp->vlru))
7318 if (idx == VLRU_QUEUE_INVALID)
7320 VLRU_Wait_r(&volume_LRU.q[idx]);
7321 } while (idx != vp->vlru.idx);
7323 /* now remove from the VLRU and update
7324 * the appropriate counter */
7325 queue_Remove(&vp->vlru);
7326 volume_LRU.q[idx].len--;
7327 vp->vlru.idx = VLRU_QUEUE_INVALID;
7328 V_attachFlags(vp) &= ~(VOL_ON_VLRU);
7332 * tell the VLRU subsystem that a volume was just accessed.
7334 * @param[in] vp pointer to volume object
7337 * @arg @c VOL_LOCK is held
7338 * @arg caller MUST hold a lightweight ref on @p vp
7339 * @arg caller MUST NOT hold exclusive ownership of any VLRU queue
7341 * @post volume VLRU access statistics are updated. If the volume was on
7342 * the VLRU soft detach candidate queue, it is moved to the NEW
7345 * @note @c VOL_LOCK may be dropped internally
7349 * @internal volume package internal use only.
7352 VLRU_UpdateAccess_r(Volume * vp)
7354 Volume * rvp = NULL;
7359 if (queue_IsNotOnQueue(&vp->vlru))
7362 opr_Assert(V_attachFlags(vp) & VOL_ON_VLRU);
7364 /* update the access timestamp */
7365 vp->stats.last_get = FT_ApproxTime();
7368 * if the volume is on the soft detach candidate
7369 * list, we need to safely move it back to a
7370 * regular generation. this has to be done
7371 * carefully so we don't race against the scanner
7375 /* if this volume is on the soft detach candidate queue,
7376 * then grab exclusive access to the necessary queues */
7377 if (vp->vlru.idx == VLRU_QUEUE_CANDIDATE) {
7379 VCreateReservation_r(rvp);
7381 VLRU_Wait_r(&volume_LRU.q[VLRU_QUEUE_NEW]);
7382 VLRU_BeginExclusive_r(&volume_LRU.q[VLRU_QUEUE_NEW]);
7383 VLRU_Wait_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7384 VLRU_BeginExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7387 /* make sure multiple threads don't race to update */
7388 if (vp->vlru.idx == VLRU_QUEUE_CANDIDATE) {
7389 VLRU_SwitchQueues(vp, VLRU_QUEUE_NEW, 1);
7393 VLRU_EndExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7394 VLRU_EndExclusive_r(&volume_LRU.q[VLRU_QUEUE_NEW]);
7395 VCancelReservation_r(rvp);
7400 * switch a volume between two VLRU queues.
7402 * @param[in] vp pointer to volume object
7403 * @param[in] new_idx index of VLRU queue onto which the volume will be moved
7404 * @param[in] append controls whether the volume will be appended or
7405 * prepended to the queue. A nonzero value means it will
7406 * be appended; zero means it will be prepended.
7408 * @pre The new (and old, if applicable) queue(s) must either be owned
7409 * exclusively by the calling thread for asynchronous manipulation,
7410 * or the queue(s) must be quiescent and VOL_LOCK must be held.
7411 * Please see VLRU_BeginExclusive_r, VLRU_EndExclusive_r and VLRU_Wait_r
7412 * for further details of the queue asynchronous processing mechanism.
7414 * @post If the volume object was already on a VLRU queue, it is
7415 * removed from the queue. Depending on the value of the append
7416 * parameter, the volume object is either appended or prepended
7417 * to the VLRU queue referenced by the new_idx parameter.
7421 * @see VLRU_BeginExclusive_r
7422 * @see VLRU_EndExclusive_r
7425 * @internal volume package internal use only.
7428 VLRU_SwitchQueues(Volume * vp, int new_idx, int append)
7430 if (queue_IsNotOnQueue(&vp->vlru))
7433 queue_Remove(&vp->vlru);
7434 volume_LRU.q[vp->vlru.idx].len--;
7436 /* put the volume back on the correct generational queue */
7438 queue_Append(&volume_LRU.q[new_idx], &vp->vlru);
7440 queue_Prepend(&volume_LRU.q[new_idx], &vp->vlru);
7443 volume_LRU.q[new_idx].len++;
7444 vp->vlru.idx = new_idx;
7448 * VLRU background thread.
7450 * The VLRU Scanner Thread is responsible for periodically scanning through
7451 * each VLRU queue looking for volumes which should be moved to another
7452 * queue, or soft detached.
7454 * @param[in] args unused thread arguments parameter
7456 * @return unused thread return value
7457 * @retval NULL always
7459 * @internal volume package internal use only.
7462 VLRU_ScannerThread(void * args)
7464 afs_uint32 now, min_delay, delay;
7465 int i, min_idx, min_op, overdue, state;
7467 /* set t=0 for promotion cycle to be
7468 * fileserver startup */
7469 now = FT_ApproxTime();
7470 for (i=0; i < VLRU_GENERATIONS-1; i++) {
7471 volume_LRU.last_promotion[i] = now;
7474 /* don't start the scanner until VLRU_offline_thresh
7475 * plus a small delay for VInitVolumePackage2 to finish
7478 sleep(VLRU_offline_thresh + 60);
7480 /* set t=0 for scan cycle to be now */
7481 now = FT_ApproxTime();
7482 for (i=0; i < VLRU_GENERATIONS+1; i++) {
7483 volume_LRU.last_scan[i] = now;
7487 if (volume_LRU.scanner_state == VLRU_SCANNER_STATE_OFFLINE) {
7488 volume_LRU.scanner_state = VLRU_SCANNER_STATE_ONLINE;
7491 while ((state = volume_LRU.scanner_state) != VLRU_SCANNER_STATE_SHUTTING_DOWN) {
7492 /* check to see if we've been asked to pause */
7493 if (volume_LRU.scanner_state == VLRU_SCANNER_STATE_PAUSING) {
7494 volume_LRU.scanner_state = VLRU_SCANNER_STATE_PAUSED;
7495 opr_cv_broadcast(&volume_LRU.cv);
7497 VOL_CV_WAIT(&volume_LRU.cv);
7498 } while (volume_LRU.scanner_state == VLRU_SCANNER_STATE_PAUSED);
7501 /* scheduling can happen outside the glock */
7504 /* figure out what is next on the schedule */
7506 /* figure out a potential schedule for the new generation first */
7508 min_delay = volume_LRU.scan_interval[0] + volume_LRU.last_scan[0] - now;
7511 if (min_delay > volume_LRU.scan_interval[0]) {
7512 /* unsigned overflow -- we're overdue to run this scan */
7517 /* if we're not overdue for gen 0, figure out schedule for candidate gen */
7519 i = VLRU_QUEUE_CANDIDATE;
7520 delay = volume_LRU.scan_interval[i] + volume_LRU.last_scan[i] - now;
7521 if (delay < min_delay) {
7525 if (delay > volume_LRU.scan_interval[i]) {
7526 /* unsigned overflow -- we're overdue to run this scan */
7533 /* if we're still not overdue for something, figure out schedules for promotions */
7534 for (i=0; !overdue && i < VLRU_GENERATIONS-1; i++) {
7535 delay = volume_LRU.promotion_interval[i] + volume_LRU.last_promotion[i] - now;
7536 if (delay < min_delay) {
7541 if (delay > volume_LRU.promotion_interval[i]) {
7542 /* unsigned overflow -- we're overdue to run this promotion */
7551 /* sleep as needed */
7556 /* do whatever is next */
7559 VLRU_Promote_r(min_idx);
7560 VLRU_Demote_r(min_idx+1);
7562 VLRU_Scan_r(min_idx);
7564 now = FT_ApproxTime();
7567 Log("VLRU scanner asked to go offline (scanner_state=%d)\n", state);
7569 /* signal that scanner is down */
7570 volume_LRU.scanner_state = VLRU_SCANNER_STATE_OFFLINE;
7571 opr_cv_broadcast(&volume_LRU.cv);
7577 * promote volumes from one VLRU generation to the next.
7579 * This routine scans a VLRU generation looking for volumes which are
7580 * eligible to be promoted to the next generation. All volumes which
7581 * meet the eligibility requirement are promoted.
7583 * Promotion eligibility is based upon meeting both of the following
7586 * @arg The volume has been accessed since the last promotion:
7587 * @c (vp->stats.last_get >= vp->stats.last_promote)
7588 * @arg The last promotion occurred at least
7589 * @c volume_LRU.promotion_interval[idx] seconds ago
7591 * As a performance optimization, promotions are "globbed". In other
7592 * words, we promote arbitrarily large contiguous sublists of elements
7595 * @param[in] idx VLRU queue index to scan
7599 * @internal VLRU internal use only.
7602 VLRU_Promote_r(int idx)
7604 int len, chaining, promote;
7605 afs_uint32 now, thresh;
7606 struct rx_queue *qp, *nqp;
7607 Volume * vp, *start = NULL, *end = NULL;
7609 /* get exclusive access to two chains, and drop the glock */
7610 VLRU_Wait_r(&volume_LRU.q[idx]);
7611 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
7612 VLRU_Wait_r(&volume_LRU.q[idx+1]);
7613 VLRU_BeginExclusive_r(&volume_LRU.q[idx+1]);
7616 thresh = volume_LRU.promotion_interval[idx];
7617 now = FT_ApproxTime();
7620 for (queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
7621 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
7622 promote = (((vp->stats.last_promote + thresh) <= now) &&
7623 (vp->stats.last_get >= vp->stats.last_promote));
7631 /* promote and prepend chain */
7632 queue_MoveChainAfter(&volume_LRU.q[idx+1], &start->vlru, &end->vlru);
7646 /* promote and prepend */
7647 queue_MoveChainAfter(&volume_LRU.q[idx+1], &start->vlru, &end->vlru);
7651 volume_LRU.q[idx].len -= len;
7652 volume_LRU.q[idx+1].len += len;
7655 /* release exclusive access to the two chains */
7657 volume_LRU.last_promotion[idx] = now;
7658 VLRU_EndExclusive_r(&volume_LRU.q[idx+1]);
7659 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
7662 /* run the demotions */
7664 VLRU_Demote_r(int idx)
7667 int len, chaining, demote;
7668 afs_uint32 now, thresh;
7669 struct rx_queue *qp, *nqp;
7670 Volume * vp, *start = NULL, *end = NULL;
7671 Volume ** salv_flag_vec = NULL;
7672 int salv_vec_offset = 0;
7674 opr_Assert(idx == VLRU_QUEUE_MID || idx == VLRU_QUEUE_OLD);
7676 /* get exclusive access to two chains, and drop the glock */
7677 VLRU_Wait_r(&volume_LRU.q[idx-1]);
7678 VLRU_BeginExclusive_r(&volume_LRU.q[idx-1]);
7679 VLRU_Wait_r(&volume_LRU.q[idx]);
7680 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
7683 /* no big deal if this allocation fails */
7684 if (volume_LRU.q[idx].len) {
7685 salv_flag_vec = malloc(volume_LRU.q[idx].len * sizeof(Volume *));
7688 now = FT_ApproxTime();
7689 thresh = volume_LRU.promotion_interval[idx-1];
7692 for (queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
7693 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
7694 demote = (((vp->stats.last_promote + thresh) <= now) &&
7695 (vp->stats.last_get < (now - thresh)));
7697 /* we now do volume update list DONT_SALVAGE flag setting during
7698 * demotion passes */
7699 if (salv_flag_vec &&
7700 !(V_attachFlags(vp) & VOL_HDR_DONTSALV) &&
7702 (vp->updateTime < (now - SALVAGE_INTERVAL)) &&
7703 (V_attachState(vp) == VOL_STATE_ATTACHED)) {
7704 salv_flag_vec[salv_vec_offset++] = vp;
7705 VCreateReservation_r(vp);
7714 /* demote and append chain */
7715 queue_MoveChainBefore(&volume_LRU.q[idx-1], &start->vlru, &end->vlru);
7729 queue_MoveChainBefore(&volume_LRU.q[idx-1], &start->vlru, &end->vlru);
7733 volume_LRU.q[idx].len -= len;
7734 volume_LRU.q[idx-1].len += len;
7737 /* release exclusive access to the two chains */
7739 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
7740 VLRU_EndExclusive_r(&volume_LRU.q[idx-1]);
7742 /* now go back and set the DONT_SALVAGE flags as appropriate */
7743 if (salv_flag_vec) {
7745 for (i = 0; i < salv_vec_offset; i++) {
7746 vp = salv_flag_vec[i];
7747 if (!(V_attachFlags(vp) & VOL_HDR_DONTSALV) &&
7748 (vp->updateTime < (now - SALVAGE_INTERVAL)) &&
7749 (V_attachState(vp) == VOL_STATE_ATTACHED)) {
7752 V_attachFlags(vp) |= VOL_HDR_DONTSALV;
7753 V_dontSalvage(vp) = DONT_SALVAGE;
7754 VUpdateVolume_r(&ec, vp, 0);
7758 VCancelReservation_r(vp);
7760 free(salv_flag_vec);
7764 /* run a pass of the VLRU GC scanner */
7766 VLRU_Scan_r(int idx)
7768 afs_uint32 now, thresh;
7769 struct rx_queue *qp, *nqp;
7773 opr_Assert(idx == VLRU_QUEUE_NEW || idx == VLRU_QUEUE_CANDIDATE);
7775 /* gain exclusive access to the idx VLRU */
7776 VLRU_Wait_r(&volume_LRU.q[idx]);
7777 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
7779 if (idx != VLRU_QUEUE_CANDIDATE) {
7780 /* gain exclusive access to the candidate VLRU */
7781 VLRU_Wait_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7782 VLRU_BeginExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7785 now = FT_ApproxTime();
7786 thresh = now - VLRU_offline_thresh;
7788 /* perform candidate selection and soft detaching */
7789 if (idx == VLRU_QUEUE_CANDIDATE) {
7790 /* soft detach some volumes from the candidate pool */
7794 for (i=0,queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
7795 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
7796 if (i >= VLRU_offline_max) {
7799 /* check timestamp to see if it's a candidate for soft detaching */
7800 if (vp->stats.last_get <= thresh) {
7802 if (VCheckSoftDetach(vp, thresh))
7808 /* scan for volumes to become soft detach candidates */
7809 for (i=1,queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue),i++) {
7810 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
7812 /* check timestamp to see if it's a candidate for soft detaching */
7813 if (vp->stats.last_get <= thresh) {
7814 VCheckSoftDetachCandidate(vp, thresh);
7817 if (!(i&0x7f)) { /* lock coarsening optimization */
7825 /* relinquish exclusive access to the VLRU chains */
7829 volume_LRU.last_scan[idx] = now;
7830 if (idx != VLRU_QUEUE_CANDIDATE) {
7831 VLRU_EndExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7833 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
7836 /* check whether volume is safe to soft detach
7837 * caller MUST NOT hold a ref count on vp */
7839 VCheckSoftDetach(Volume * vp, afs_uint32 thresh)
7843 if (vp->nUsers || vp->nWaiters)
7846 if (vp->stats.last_get <= thresh) {
7847 ret = VSoftDetachVolume_r(vp, thresh);
7853 /* check whether volume should be made a
7854 * soft detach candidate */
7856 VCheckSoftDetachCandidate(Volume * vp, afs_uint32 thresh)
7859 if (vp->nUsers || vp->nWaiters)
7864 opr_Assert(idx == VLRU_QUEUE_NEW);
7866 if (vp->stats.last_get <= thresh) {
7867 /* move to candidate pool */
7868 queue_Remove(&vp->vlru);
7869 volume_LRU.q[VLRU_QUEUE_NEW].len--;
7870 queue_Prepend(&volume_LRU.q[VLRU_QUEUE_CANDIDATE], &vp->vlru);
7871 vp->vlru.idx = VLRU_QUEUE_CANDIDATE;
7872 volume_LRU.q[VLRU_QUEUE_CANDIDATE].len++;
7880 /* begin exclusive access on VLRU */
7882 VLRU_BeginExclusive_r(struct VLRU_q * q)
7884 opr_Assert(q->busy == 0);
7888 /* end exclusive access on VLRU */
7890 VLRU_EndExclusive_r(struct VLRU_q * q)
7892 opr_Assert(q->busy);
7894 opr_cv_broadcast(&q->cv);
7897 /* wait for another thread to end exclusive access on VLRU */
7899 VLRU_Wait_r(struct VLRU_q * q)
7902 VOL_CV_WAIT(&q->cv);
7907 * volume soft detach
7909 * caller MUST NOT hold a ref count on vp */
7911 VSoftDetachVolume_r(Volume * vp, afs_uint32 thresh)
7916 opr_Assert(vp->vlru.idx == VLRU_QUEUE_CANDIDATE);
7918 ts_save = vp->stats.last_get;
7919 if (ts_save > thresh)
7922 if (vp->nUsers || vp->nWaiters)
7925 if (VIsExclusiveState(V_attachState(vp))) {
7929 switch (V_attachState(vp)) {
7930 case VOL_STATE_UNATTACHED:
7931 case VOL_STATE_PREATTACHED:
7932 case VOL_STATE_ERROR:
7933 case VOL_STATE_GOING_OFFLINE:
7934 case VOL_STATE_SHUTTING_DOWN:
7935 case VOL_STATE_SALVAGING:
7936 case VOL_STATE_DELETED:
7937 volume_LRU.q[vp->vlru.idx].len--;
7939 /* create and cancel a reservation to
7940 * give the volume an opportunity to
7942 VCreateReservation_r(vp);
7943 queue_Remove(&vp->vlru);
7944 vp->vlru.idx = VLRU_QUEUE_INVALID;
7945 V_attachFlags(vp) &= ~(VOL_ON_VLRU);
7946 VCancelReservation_r(vp);
7952 /* hold the volume and take it offline.
7953 * no need for reservations, as VHold_r
7954 * takes care of that internally. */
7955 if (VHold_r(vp) == 0) {
7956 /* vhold drops the glock, so now we should
7957 * check to make sure we aren't racing against
7958 * other threads. if we are racing, offlining vp
7959 * would be wasteful, and block the scanner for a while
7963 (vp->shuttingDown) ||
7964 (vp->goingOffline) ||
7965 (vp->stats.last_get != ts_save)) {
7966 /* looks like we're racing someone else. bail */
7970 /* pull it off the VLRU */
7971 opr_Assert(vp->vlru.idx == VLRU_QUEUE_CANDIDATE);
7972 volume_LRU.q[VLRU_QUEUE_CANDIDATE].len--;
7973 queue_Remove(&vp->vlru);
7974 vp->vlru.idx = VLRU_QUEUE_INVALID;
7975 V_attachFlags(vp) &= ~(VOL_ON_VLRU);
7977 /* take if offline */
7978 VOffline_r(vp, "volume has been soft detached");
7980 /* invalidate the volume header cache */
7981 FreeVolumeHeader(vp);
7984 IncUInt64(&VStats.soft_detaches);
7985 vp->stats.soft_detaches++;
7987 /* put in pre-attached state so demand
7988 * attacher can work on it */
7989 VChangeState_r(vp, VOL_STATE_PREATTACHED);
7995 #endif /* AFS_DEMAND_ATTACH_FS */
7998 /***************************************************/
7999 /* Volume Header Cache routines */
8000 /***************************************************/
8003 * volume header cache.
8005 struct volume_hdr_LRU_t volume_hdr_LRU;
8008 * initialize the volume header cache.
8010 * @param[in] howMany number of header cache entries to preallocate
8012 * @pre VOL_LOCK held. Function has never been called before.
8014 * @post howMany cache entries are allocated, initialized, and added
8015 * to the LRU list. Header cache statistics are initialized.
8017 * @note only applicable to fileServer program type. Should only be
8018 * called once during volume package initialization.
8020 * @internal volume package internal use only.
8023 VInitVolumeHeaderCache(afs_uint32 howMany)
8025 struct volHeader *hp;
8026 if (programType != fileServer)
8028 queue_Init(&volume_hdr_LRU);
8029 volume_hdr_LRU.stats.free = 0;
8030 volume_hdr_LRU.stats.used = howMany;
8031 volume_hdr_LRU.stats.attached = 0;
8032 hp = (struct volHeader *)(calloc(howMany, sizeof(struct volHeader)));
8033 opr_Assert(hp != NULL);
8036 /* We are using ReleaseVolumeHeader to initialize the values on the header list
8037 * to ensure they have the right values
8039 ReleaseVolumeHeader(hp++);
8042 /* get a volume header off of the volume header LRU.
8044 * @return volume header
8045 * @retval NULL no usable volume header is available on the LRU
8047 * @pre VOL_LOCK held
8049 * @post for DAFS, if the returned header is associated with a volume, that
8050 * volume is NOT in an exclusive state
8052 * @internal volume package internal use only.
8054 #ifdef AFS_DEMAND_ATTACH_FS
8055 static struct volHeader*
8056 GetVolHeaderFromLRU(void)
8058 struct volHeader *hd = NULL, *qh, *nqh;
8059 /* Usually, a volume in an exclusive state will not have its header on
8060 * the LRU. However, it is possible for this to occur when a salvage
8061 * request is received over FSSYNC, and possibly in other corner cases.
8062 * So just skip over headers whose volumes are in an exclusive state. We
8063 * could VWaitExclusiveState_r instead, but not waiting is faster and
8065 for (queue_Scan(&volume_hdr_LRU, qh, nqh, volHeader)) {
8066 if (!qh->back || !VIsExclusiveState(V_attachState(qh->back))) {
8074 #else /* AFS_DEMAND_ATTACH_FS */
8075 static struct volHeader*
8076 GetVolHeaderFromLRU(void)
8078 struct volHeader *hd = NULL;
8079 if (queue_IsNotEmpty(&volume_hdr_LRU)) {
8080 hd = queue_First(&volume_hdr_LRU, volHeader);
8085 #endif /* !AFS_DEMAND_ATTACH_FS */
8088 * get a volume header and attach it to the volume object.
8090 * @param[in] vp pointer to volume object
8092 * @return cache entry status
8093 * @retval 0 volume header was newly attached; cache data is invalid
8094 * @retval 1 volume header was previously attached; cache data is valid
8096 * @pre VOL_LOCK held. For DAFS, lightweight ref must be held on volume object.
8098 * @post volume header attached to volume object. if necessary, header cache
8099 * entry on LRU is synchronized to disk. Header is removed from LRU list.
8101 * @note VOL_LOCK may be dropped
8103 * @warning this interface does not load header data from disk. it merely
8104 * attaches a header object to the volume object, and may sync the old
8105 * header cache data out to disk in the process.
8107 * @internal volume package internal use only.
8110 GetVolumeHeader(Volume * vp)
8113 struct volHeader *hd;
8115 static int everLogged = 0;
8117 #ifdef AFS_DEMAND_ATTACH_FS
8118 VolState vp_save = 0, back_save = 0;
8120 /* XXX debug 9/19/05 we've apparently got
8121 * a ref counting bug somewhere that's
8122 * breaking the nUsers == 0 => header on LRU
8124 if (vp->header && queue_IsNotOnQueue(vp->header)) {
8125 Log("nUsers == 0, but header not on LRU\n");
8130 old = (vp->header != NULL); /* old == volume already has a header */
8132 if (programType != fileServer) {
8133 /* for volume utilities, we allocate volHeaders as needed */
8135 hd = calloc(1, sizeof(*vp->header));
8136 opr_Assert(hd != NULL);
8139 #ifdef AFS_DEMAND_ATTACH_FS
8140 V_attachFlags(vp) |= VOL_HDR_ATTACHED;
8144 /* for the fileserver, we keep a volume header cache */
8146 /* the header we previously dropped in the lru is
8147 * still available. pull it off the lru and return */
8150 opr_Assert(hd->back == vp);
8151 #ifdef AFS_DEMAND_ATTACH_FS
8152 V_attachFlags(vp) &= ~(VOL_HDR_IN_LRU);
8155 hd = GetVolHeaderFromLRU();
8157 /* LRU is empty, so allocate a new volHeader
8158 * this is probably indicative of a leak, so let the user know */
8159 hd = calloc(1, sizeof(struct volHeader));
8160 opr_Assert(hd != NULL);
8162 Log("****Allocated more volume headers, probably leak****\n");
8165 volume_hdr_LRU.stats.free++;
8168 /* this header used to belong to someone else.
8169 * we'll need to check if the header needs to
8170 * be sync'd out to disk */
8172 #ifdef AFS_DEMAND_ATTACH_FS
8173 /* GetVolHeaderFromLRU had better not give us back a header
8174 * with a volume in exclusive state... */
8175 opr_Assert(!VIsExclusiveState(V_attachState(hd->back)));
8178 if (hd->diskstuff.inUse) {
8179 /* volume was in use, so we'll need to sync
8180 * its header to disk */
8182 #ifdef AFS_DEMAND_ATTACH_FS
8183 back_save = VChangeState_r(hd->back, VOL_STATE_UPDATING);
8184 vp_save = VChangeState_r(vp, VOL_STATE_HDR_ATTACHING);
8185 VCreateReservation_r(hd->back);
8189 WriteVolumeHeader_r(&error, hd->back);
8190 /* Ignore errors; catch them later */
8192 #ifdef AFS_DEMAND_ATTACH_FS
8197 hd->back->header = NULL;
8198 #ifdef AFS_DEMAND_ATTACH_FS
8199 V_attachFlags(hd->back) &= ~(VOL_HDR_ATTACHED | VOL_HDR_LOADED | VOL_HDR_IN_LRU);
8201 if (hd->diskstuff.inUse) {
8202 VChangeState_r(hd->back, back_save);
8203 VCancelReservation_r(hd->back);
8204 VChangeState_r(vp, vp_save);
8208 volume_hdr_LRU.stats.attached++;
8212 #ifdef AFS_DEMAND_ATTACH_FS
8213 V_attachFlags(vp) |= VOL_HDR_ATTACHED;
8216 volume_hdr_LRU.stats.free--;
8217 volume_hdr_LRU.stats.used++;
8219 IncUInt64(&VStats.hdr_gets);
8220 #ifdef AFS_DEMAND_ATTACH_FS
8221 IncUInt64(&vp->stats.hdr_gets);
8222 vp->stats.last_hdr_get = FT_ApproxTime();
8229 * make sure volume header is attached and contains valid cache data.
8231 * @param[out] ec outbound error code
8232 * @param[in] vp pointer to volume object
8234 * @pre VOL_LOCK held. For DAFS, lightweight ref held on vp.
8236 * @post header cache entry attached, and loaded with valid data, or
8237 * *ec is nonzero, and the header is released back into the LRU.
8239 * @internal volume package internal use only.
8242 LoadVolumeHeader(Error * ec, Volume * vp)
8244 #ifdef AFS_DEMAND_ATTACH_FS
8245 VolState state_save;
8249 if (vp->nUsers == 0 && !GetVolumeHeader(vp)) {
8250 IncUInt64(&VStats.hdr_loads);
8251 state_save = VChangeState_r(vp, VOL_STATE_HDR_LOADING);
8254 ReadHeader(ec, V_diskDataHandle(vp), (char *)&V_disk(vp),
8255 sizeof(V_disk(vp)), VOLUMEINFOMAGIC,
8257 IncUInt64(&vp->stats.hdr_loads);
8258 now = FT_ApproxTime();
8262 V_attachFlags(vp) |= VOL_HDR_LOADED;
8263 vp->stats.last_hdr_load = now;
8265 VChangeState_r(vp, state_save);
8267 #else /* AFS_DEMAND_ATTACH_FS */
8269 if (vp->nUsers == 0 && !GetVolumeHeader(vp)) {
8270 IncUInt64(&VStats.hdr_loads);
8272 ReadHeader(ec, V_diskDataHandle(vp), (char *)&V_disk(vp),
8273 sizeof(V_disk(vp)), VOLUMEINFOMAGIC,
8276 #endif /* AFS_DEMAND_ATTACH_FS */
8278 /* maintain (nUsers==0) => header in LRU invariant */
8279 FreeVolumeHeader(vp);
8284 * release a header cache entry back into the LRU list.
8286 * @param[in] hd pointer to volume header cache object
8288 * @pre VOL_LOCK held.
8290 * @post header cache object appended onto end of LRU list.
8292 * @note only applicable to fileServer program type.
8294 * @note used to place a header cache entry back into the
8295 * LRU pool without invalidating it as a cache entry.
8297 * @internal volume package internal use only.
8300 ReleaseVolumeHeader(struct volHeader *hd)
8302 if (programType != fileServer)
8304 if (!hd || queue_IsOnQueue(hd)) /* no header, or header already released */
8306 queue_Append(&volume_hdr_LRU, hd);
8307 #ifdef AFS_DEMAND_ATTACH_FS
8309 V_attachFlags(hd->back) |= VOL_HDR_IN_LRU;
8312 volume_hdr_LRU.stats.free++;
8313 volume_hdr_LRU.stats.used--;
8317 * free/invalidate a volume header cache entry.
8319 * @param[in] vp pointer to volume object
8321 * @pre VOL_LOCK is held.
8323 * @post For fileserver, header cache entry is returned to LRU, and it is
8324 * invalidated as a cache entry. For volume utilities, the header
8325 * cache entry is freed.
8327 * @note For fileserver, this should be utilized instead of ReleaseVolumeHeader
8328 * whenever it is necessary to invalidate the header cache entry.
8330 * @see ReleaseVolumeHeader
8332 * @internal volume package internal use only.
8335 FreeVolumeHeader(Volume * vp)
8337 struct volHeader *hd = vp->header;
8340 if (programType == fileServer) {
8341 ReleaseVolumeHeader(hd);
8346 #ifdef AFS_DEMAND_ATTACH_FS
8347 V_attachFlags(vp) &= ~(VOL_HDR_ATTACHED | VOL_HDR_IN_LRU | VOL_HDR_LOADED);
8349 volume_hdr_LRU.stats.attached--;
8354 /***************************************************/
8355 /* Volume Hash Table routines */
8356 /***************************************************/
8359 * set size of volume object hash table.
8361 * @param[in] logsize log(2) of desired hash table size
8363 * @return operation status
8365 * @retval -1 failure
8367 * @pre MUST be called prior to VInitVolumePackage2
8369 * @post Volume Hash Table will have 2^logsize buckets
8372 VSetVolHashSize(int logsize)
8374 /* 64 to 268435456 hash buckets seems like a reasonable range */
8375 if ((logsize < 6 ) || (logsize > 28)) {
8380 VolumeHashTable.Size = 1 << logsize;
8381 VolumeHashTable.Mask = VolumeHashTable.Size - 1;
8383 /* we can't yet support runtime modification of this
8384 * parameter. we'll need a configuration rwlock to
8385 * make runtime modification feasible.... */
8392 * initialize dynamic data structures for volume hash table.
8394 * @post hash table is allocated, and fields are initialized.
8396 * @internal volume package internal use only.
8399 VInitVolumeHash(void)
8403 VolumeHashTable.Table = (VolumeHashChainHead *) calloc(VolumeHashTable.Size,
8404 sizeof(VolumeHashChainHead));
8405 opr_Assert(VolumeHashTable.Table != NULL);
8407 for (i=0; i < VolumeHashTable.Size; i++) {
8408 queue_Init(&VolumeHashTable.Table[i]);
8409 #ifdef AFS_DEMAND_ATTACH_FS
8410 opr_cv_init(&VolumeHashTable.Table[i].chain_busy_cv);
8411 #endif /* AFS_DEMAND_ATTACH_FS */
8416 * add a volume object to the hash table.
8418 * @param[in] vp pointer to volume object
8419 * @param[in] hashid hash of volume id
8421 * @pre VOL_LOCK is held. For DAFS, caller must hold a lightweight
8424 * @post volume is added to hash chain.
8426 * @internal volume package internal use only.
8428 * @note For DAFS, VOL_LOCK may be dropped in order to wait for an
8429 * asynchronous hash chain reordering to finish.
8432 AddVolumeToHashTable(Volume * vp, VolumeId hashid)
8434 VolumeHashChainHead * head;
8436 if (queue_IsOnQueue(vp))
8439 head = &VolumeHashTable.Table[VOLUME_HASH(hashid)];
8441 #ifdef AFS_DEMAND_ATTACH_FS
8442 /* wait for the hash chain to become available */
8445 V_attachFlags(vp) |= VOL_IN_HASH;
8446 vp->chainCacheCheck = ++head->cacheCheck;
8447 #endif /* AFS_DEMAND_ATTACH_FS */
8450 vp->hashid = hashid;
8451 queue_Append(head, vp);
8452 vp->vnodeHashOffset = VolumeHashOffset_r();
8456 * delete a volume object from the hash table.
8458 * @param[in] vp pointer to volume object
8460 * @pre VOL_LOCK is held. For DAFS, caller must hold a lightweight
8463 * @post volume is removed from hash chain.
8465 * @internal volume package internal use only.
8467 * @note For DAFS, VOL_LOCK may be dropped in order to wait for an
8468 * asynchronous hash chain reordering to finish.
8471 DeleteVolumeFromHashTable(Volume * vp)
8473 VolumeHashChainHead * head;
8475 if (!queue_IsOnQueue(vp))
8478 head = &VolumeHashTable.Table[VOLUME_HASH(vp->hashid)];
8480 #ifdef AFS_DEMAND_ATTACH_FS
8481 /* wait for the hash chain to become available */
8484 V_attachFlags(vp) &= ~(VOL_IN_HASH);
8486 #endif /* AFS_DEMAND_ATTACH_FS */
8490 /* do NOT reset hashid to zero, as the online
8491 * salvager package may need to know the volume id
8492 * after the volume is removed from the hash */
8496 * lookup a volume object in the hash table given a volume id.
8498 * @param[out] ec error code return
8499 * @param[in] volumeId volume id
8500 * @param[in] hint volume object which we believe could be the correct
8503 * @return volume object pointer
8504 * @retval NULL no such volume id is registered with the hash table.
8506 * @pre VOL_LOCK is held. For DAFS, caller must hold a lightweight
8509 * @post volume object with the given id is returned. volume object and
8510 * hash chain access statistics are updated. hash chain may have
8513 * @note For DAFS, VOL_LOCK may be dropped in order to wait for an
8514 * asynchronous hash chain reordering operation to finish, or
8515 * in order for us to perform an asynchronous chain reordering.
8517 * @note Hash chain reorderings occur when the access count for the
8518 * volume object being looked up exceeds the sum of the previous
8519 * node's (the node ahead of it in the hash chain linked list)
8520 * access count plus the constant VOLUME_HASH_REORDER_THRESHOLD.
8522 * @note For DAFS, the hint parameter allows us to short-circuit if the
8523 * cacheCheck fields match between the hash chain head and the
8524 * hint volume object.
8527 VLookupVolume_r(Error * ec, VolumeId volumeId, Volume * hint)
8531 #ifdef AFS_DEMAND_ATTACH_FS
8534 VolumeHashChainHead * head;
8537 head = &VolumeHashTable.Table[VOLUME_HASH(volumeId)];
8539 #ifdef AFS_DEMAND_ATTACH_FS
8540 /* wait for the hash chain to become available */
8543 /* check to see if we can short circuit without walking the hash chain */
8544 if (hint && (hint->chainCacheCheck == head->cacheCheck)) {
8545 IncUInt64(&hint->stats.hash_short_circuits);
8548 #endif /* AFS_DEMAND_ATTACH_FS */
8550 /* someday we need to either do per-chain locks, RWlocks,
8551 * or both for volhash access.
8552 * (and move to a data structure with better cache locality) */
8554 /* search the chain for this volume id */
8555 for(queue_Scan(head, vp, np, Volume)) {
8557 if (vp->hashid == volumeId) {
8562 if (queue_IsEnd(head, vp)) {
8566 #ifdef AFS_DEMAND_ATTACH_FS
8567 /* update hash chain statistics */
8570 FillInt64(lks, 0, looks);
8571 AddUInt64(head->looks, lks, &head->looks);
8572 AddUInt64(VStats.hash_looks, lks, &VStats.hash_looks);
8573 IncUInt64(&head->gets);
8578 IncUInt64(&vp->stats.hash_lookups);
8580 /* for demand attach fileserver, we permit occasional hash chain reordering
8581 * so that frequently looked up volumes move towards the head of the chain */
8582 pp = queue_Prev(vp, Volume);
8583 if (!queue_IsEnd(head, pp)) {
8584 FillInt64(thresh, 0, VOLUME_HASH_REORDER_THRESHOLD);
8585 AddUInt64(thresh, pp->stats.hash_lookups, &thresh);
8586 if (GEInt64(vp->stats.hash_lookups, thresh)) {
8587 VReorderHash_r(head, pp, vp);
8591 /* update the short-circuit cache check */
8592 vp->chainCacheCheck = head->cacheCheck;
8594 #endif /* AFS_DEMAND_ATTACH_FS */
8599 #ifdef AFS_DEMAND_ATTACH_FS
8600 /* perform volume hash chain reordering.
8602 * advance a subchain beginning at vp ahead of
8603 * the adjacent subchain ending at pp */
8605 VReorderHash_r(VolumeHashChainHead * head, Volume * pp, Volume * vp)
8607 Volume *tp, *np, *lp;
8608 afs_uint64 move_thresh;
8610 /* this should never be called if the chain is already busy, so
8611 * no need to wait for other exclusive chain ops to finish */
8613 /* this is a rather heavy set of operations,
8614 * so let's set the chain busy flag and drop
8616 VHashBeginExclusive_r(head);
8619 /* scan forward in the chain from vp looking for the last element
8620 * in the chain we want to advance */
8621 FillInt64(move_thresh, 0, VOLUME_HASH_REORDER_CHAIN_THRESH);
8622 AddUInt64(move_thresh, pp->stats.hash_lookups, &move_thresh);
8623 for(queue_ScanFrom(head, vp, tp, np, Volume)) {
8624 if (LTInt64(tp->stats.hash_lookups, move_thresh)) {
8628 lp = queue_Prev(tp, Volume);
8630 /* scan backwards from pp to determine where to splice and
8631 * insert the subchain we're advancing */
8632 for(queue_ScanBackwardsFrom(head, pp, tp, np, Volume)) {
8633 if (GTInt64(tp->stats.hash_lookups, move_thresh)) {
8637 tp = queue_Next(tp, Volume);
8639 /* rebalance chain(vp,...,lp) ahead of chain(tp,...,pp) */
8640 queue_MoveChainBefore(tp,vp,lp);
8643 IncUInt64(&VStats.hash_reorders);
8645 IncUInt64(&head->reorders);
8647 /* wake up any threads waiting for the hash chain */
8648 VHashEndExclusive_r(head);
8652 /* demand-attach fs volume hash
8653 * asynchronous exclusive operations */
8656 * begin an asynchronous exclusive operation on a volume hash chain.
8658 * @param[in] head pointer to volume hash chain head object
8660 * @pre VOL_LOCK held. hash chain is quiescent.
8662 * @post hash chain marked busy.
8664 * @note this interface is used in conjunction with VHashEndExclusive_r and
8665 * VHashWait_r to perform asynchronous (wrt VOL_LOCK) operations on a
8666 * volume hash chain. Its main use case is hash chain reordering, which
8667 * has the potential to be a highly latent operation.
8669 * @see VHashEndExclusive_r
8674 * @internal volume package internal use only.
8677 VHashBeginExclusive_r(VolumeHashChainHead * head)
8679 opr_Assert(head->busy == 0);
8684 * relinquish exclusive ownership of a volume hash chain.
8686 * @param[in] head pointer to volume hash chain head object
8688 * @pre VOL_LOCK held. thread owns the hash chain exclusively.
8690 * @post hash chain is marked quiescent. threads awaiting use of
8691 * chain are awakened.
8693 * @see VHashBeginExclusive_r
8698 * @internal volume package internal use only.
8701 VHashEndExclusive_r(VolumeHashChainHead * head)
8703 opr_Assert(head->busy);
8705 opr_cv_broadcast(&head->chain_busy_cv);
8709 * wait for all asynchronous operations on a hash chain to complete.
8711 * @param[in] head pointer to volume hash chain head object
8713 * @pre VOL_LOCK held.
8715 * @post hash chain object is quiescent.
8717 * @see VHashBeginExclusive_r
8718 * @see VHashEndExclusive_r
8722 * @note This interface should be called before any attempt to
8723 * traverse the hash chain. It is permissible for a thread
8724 * to gain exclusive access to the chain, and then perform
8725 * latent operations on the chain asynchronously wrt the
8728 * @warning if waiting is necessary, VOL_LOCK is dropped
8730 * @internal volume package internal use only.
8733 VHashWait_r(VolumeHashChainHead * head)
8735 while (head->busy) {
8736 VOL_CV_WAIT(&head->chain_busy_cv);
8739 #endif /* AFS_DEMAND_ATTACH_FS */
8742 /***************************************************/
8743 /* Volume by Partition List routines */
8744 /***************************************************/
8747 * demand attach fileserver adds a
8748 * linked list of volumes to each
8749 * partition object, thus allowing
8750 * for quick enumeration of all
8751 * volumes on a partition
8754 #ifdef AFS_DEMAND_ATTACH_FS
8756 * add a volume to its disk partition VByPList.
8758 * @param[in] vp pointer to volume object
8760 * @pre either the disk partition VByPList is owned exclusively
8761 * by the calling thread, or the list is quiescent and
8764 * @post volume is added to disk partition VByPList
8768 * @warning it is the caller's responsibility to ensure list
8771 * @see VVByPListWait_r
8772 * @see VVByPListBeginExclusive_r
8773 * @see VVByPListEndExclusive_r
8775 * @internal volume package internal use only.
8778 AddVolumeToVByPList_r(Volume * vp)
8780 if (queue_IsNotOnQueue(&vp->vol_list)) {
8781 queue_Append(&vp->partition->vol_list, &vp->vol_list);
8782 V_attachFlags(vp) |= VOL_ON_VBYP_LIST;
8783 vp->partition->vol_list.len++;
8788 * delete a volume from its disk partition VByPList.
8790 * @param[in] vp pointer to volume object
8792 * @pre either the disk partition VByPList is owned exclusively
8793 * by the calling thread, or the list is quiescent and
8796 * @post volume is removed from the disk partition VByPList
8800 * @warning it is the caller's responsibility to ensure list
8803 * @see VVByPListWait_r
8804 * @see VVByPListBeginExclusive_r
8805 * @see VVByPListEndExclusive_r
8807 * @internal volume package internal use only.
8810 DeleteVolumeFromVByPList_r(Volume * vp)
8812 if (queue_IsOnQueue(&vp->vol_list)) {
8813 queue_Remove(&vp->vol_list);
8814 V_attachFlags(vp) &= ~(VOL_ON_VBYP_LIST);
8815 vp->partition->vol_list.len--;
8820 * begin an asynchronous exclusive operation on a VByPList.
8822 * @param[in] dp pointer to disk partition object
8824 * @pre VOL_LOCK held. VByPList is quiescent.
8826 * @post VByPList marked busy.
8828 * @note this interface is used in conjunction with VVByPListEndExclusive_r and
8829 * VVByPListWait_r to perform asynchronous (wrt VOL_LOCK) operations on a
8832 * @see VVByPListEndExclusive_r
8833 * @see VVByPListWait_r
8837 * @internal volume package internal use only.
8839 /* take exclusive control over the list */
8841 VVByPListBeginExclusive_r(struct DiskPartition64 * dp)
8843 opr_Assert(dp->vol_list.busy == 0);
8844 dp->vol_list.busy = 1;
8848 * relinquish exclusive ownership of a VByPList.
8850 * @param[in] dp pointer to disk partition object
8852 * @pre VOL_LOCK held. thread owns the VByPList exclusively.
8854 * @post VByPList is marked quiescent. threads awaiting use of
8855 * the list are awakened.
8857 * @see VVByPListBeginExclusive_r
8858 * @see VVByPListWait_r
8862 * @internal volume package internal use only.
8865 VVByPListEndExclusive_r(struct DiskPartition64 * dp)
8867 opr_Assert(dp->vol_list.busy);
8868 dp->vol_list.busy = 0;
8869 opr_cv_broadcast(&dp->vol_list.cv);
8873 * wait for all asynchronous operations on a VByPList to complete.
8875 * @param[in] dp pointer to disk partition object
8877 * @pre VOL_LOCK is held.
8879 * @post disk partition's VByP list is quiescent
8883 * @note This interface should be called before any attempt to
8884 * traverse the VByPList. It is permissible for a thread
8885 * to gain exclusive access to the list, and then perform
8886 * latent operations on the list asynchronously wrt the
8889 * @warning if waiting is necessary, VOL_LOCK is dropped
8891 * @see VVByPListEndExclusive_r
8892 * @see VVByPListBeginExclusive_r
8894 * @internal volume package internal use only.
8897 VVByPListWait_r(struct DiskPartition64 * dp)
8899 while (dp->vol_list.busy) {
8900 VOL_CV_WAIT(&dp->vol_list.cv);
8903 #endif /* AFS_DEMAND_ATTACH_FS */
8905 /***************************************************/
8906 /* Volume Cache Statistics routines */
8907 /***************************************************/
8910 VPrintCacheStats_r(void)
8912 struct VnodeClassInfo *vcp;
8913 vcp = &VnodeClassInfo[vLarge];
8914 Log("Large vnode cache, %d entries, %d allocs, %d gets (%d reads), %d writes\n", vcp->cacheSize, vcp->allocs, vcp->gets, vcp->reads, vcp->writes);
8915 vcp = &VnodeClassInfo[vSmall];
8916 Log("Small vnode cache,%d entries, %d allocs, %d gets (%d reads), %d writes\n", vcp->cacheSize, vcp->allocs, vcp->gets, vcp->reads, vcp->writes);
8917 Log("Volume header cache, %d entries, %"AFS_INT64_FMT" gets, "
8918 "%"AFS_INT64_FMT" replacements\n",
8919 VStats.hdr_cache_size, VStats.hdr_gets, VStats.hdr_loads);
8923 VPrintCacheStats(void)
8926 VPrintCacheStats_r();
8930 #ifdef AFS_DEMAND_ATTACH_FS
8932 UInt64ToDouble(afs_uint64 * x)
8934 static double c32 = 4.0 * 1.073741824 * 1000000000.0;
8936 SplitInt64(*x, h, l);
8937 return (((double)h) * c32) + ((double) l);
8941 DoubleToPrintable(double x, char * buf, int len)
8943 static double billion = 1000000000.0;
8946 y[0] = (afs_uint32) (x / (billion * billion));
8947 y[1] = (afs_uint32) ((x - (((double)y[0]) * billion * billion)) / billion);
8948 y[2] = (afs_uint32) (x - ((((double)y[0]) * billion * billion) + (((double)y[1]) * billion)));
8951 snprintf(buf, len, "%d%09d%09d", y[0], y[1], y[2]);
8953 snprintf(buf, len, "%d%09d", y[1], y[2]);
8955 snprintf(buf, len, "%d", y[2]);
8961 struct VLRUExtStatsEntry {
8965 struct VLRUExtStats {
8971 } queue_info[VLRU_QUEUE_INVALID];
8972 struct VLRUExtStatsEntry * vec;
8976 * add a 256-entry fudge factor onto the vector in case state changes
8977 * out from under us.
8979 #define VLRU_EXT_STATS_VEC_LEN_FUDGE 256
8982 * collect extended statistics for the VLRU subsystem.
8984 * @param[out] stats pointer to stats structure to be populated
8985 * @param[in] nvols number of volumes currently known to exist
8987 * @pre VOL_LOCK held
8989 * @post stats->vec allocated and populated
8991 * @return operation status
8996 VVLRUExtStats_r(struct VLRUExtStats * stats, afs_uint32 nvols)
8998 afs_uint32 cur, idx, len;
8999 struct rx_queue * qp, * nqp;
9001 struct VLRUExtStatsEntry * vec;
9003 len = nvols + VLRU_EXT_STATS_VEC_LEN_FUDGE;
9004 vec = stats->vec = calloc(len,
9005 sizeof(struct VLRUExtStatsEntry));
9011 for (idx = VLRU_QUEUE_NEW; idx < VLRU_QUEUE_INVALID; idx++) {
9012 VLRU_Wait_r(&volume_LRU.q[idx]);
9013 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
9016 stats->queue_info[idx].start = cur;
9018 for (queue_Scan(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
9020 /* out of space in vec */
9023 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
9024 vec[cur].volid = vp->hashid;
9028 stats->queue_info[idx].len = cur - stats->queue_info[idx].start;
9031 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
9039 #define ENUMTOSTRING(en) #en
9040 #define ENUMCASE(en) \
9041 case en: return ENUMTOSTRING(en)
9044 vlru_idx_to_string(int idx)
9047 ENUMCASE(VLRU_QUEUE_NEW);
9048 ENUMCASE(VLRU_QUEUE_MID);
9049 ENUMCASE(VLRU_QUEUE_OLD);
9050 ENUMCASE(VLRU_QUEUE_CANDIDATE);
9051 ENUMCASE(VLRU_QUEUE_HELD);
9052 ENUMCASE(VLRU_QUEUE_INVALID);
9054 return "**UNKNOWN**";
9059 VPrintExtendedCacheStats_r(int flags)
9062 afs_uint32 vol_sum = 0;
9069 struct stats looks, gets, reorders, len;
9070 struct stats ch_looks, ch_gets, ch_reorders;
9072 VolumeHashChainHead *head;
9074 struct VLRUExtStats vlru_stats;
9076 /* zero out stats */
9077 memset(&looks, 0, sizeof(struct stats));
9078 memset(&gets, 0, sizeof(struct stats));
9079 memset(&reorders, 0, sizeof(struct stats));
9080 memset(&len, 0, sizeof(struct stats));
9081 memset(&ch_looks, 0, sizeof(struct stats));
9082 memset(&ch_gets, 0, sizeof(struct stats));
9083 memset(&ch_reorders, 0, sizeof(struct stats));
9085 for (i = 0; i < VolumeHashTable.Size; i++) {
9086 head = &VolumeHashTable.Table[i];
9089 VHashBeginExclusive_r(head);
9092 ch_looks.sum = UInt64ToDouble(&head->looks);
9093 ch_gets.sum = UInt64ToDouble(&head->gets);
9094 ch_reorders.sum = UInt64ToDouble(&head->reorders);
9096 /* update global statistics */
9098 looks.sum += ch_looks.sum;
9099 gets.sum += ch_gets.sum;
9100 reorders.sum += ch_reorders.sum;
9101 len.sum += (double)head->len;
9102 vol_sum += head->len;
9105 len.min = (double) head->len;
9106 len.max = (double) head->len;
9107 looks.min = ch_looks.sum;
9108 looks.max = ch_looks.sum;
9109 gets.min = ch_gets.sum;
9110 gets.max = ch_gets.sum;
9111 reorders.min = ch_reorders.sum;
9112 reorders.max = ch_reorders.sum;
9114 if (((double)head->len) < len.min)
9115 len.min = (double) head->len;
9116 if (((double)head->len) > len.max)
9117 len.max = (double) head->len;
9118 if (ch_looks.sum < looks.min)
9119 looks.min = ch_looks.sum;
9120 else if (ch_looks.sum > looks.max)
9121 looks.max = ch_looks.sum;
9122 if (ch_gets.sum < gets.min)
9123 gets.min = ch_gets.sum;
9124 else if (ch_gets.sum > gets.max)
9125 gets.max = ch_gets.sum;
9126 if (ch_reorders.sum < reorders.min)
9127 reorders.min = ch_reorders.sum;
9128 else if (ch_reorders.sum > reorders.max)
9129 reorders.max = ch_reorders.sum;
9133 if ((flags & VOL_STATS_PER_CHAIN2) && queue_IsNotEmpty(head)) {
9134 /* compute detailed per-chain stats */
9135 struct stats hdr_loads, hdr_gets;
9136 double v_looks, v_loads, v_gets;
9138 /* initialize stats with data from first element in chain */
9139 vp = queue_First(head, Volume);
9140 v_looks = UInt64ToDouble(&vp->stats.hash_lookups);
9141 v_loads = UInt64ToDouble(&vp->stats.hdr_loads);
9142 v_gets = UInt64ToDouble(&vp->stats.hdr_gets);
9143 ch_gets.min = ch_gets.max = v_looks;
9144 hdr_loads.min = hdr_loads.max = v_loads;
9145 hdr_gets.min = hdr_gets.max = v_gets;
9146 hdr_loads.sum = hdr_gets.sum = 0;
9148 vp = queue_Next(vp, Volume);
9150 /* pull in stats from remaining elements in chain */
9151 for (queue_ScanFrom(head, vp, vp, np, Volume)) {
9152 v_looks = UInt64ToDouble(&vp->stats.hash_lookups);
9153 v_loads = UInt64ToDouble(&vp->stats.hdr_loads);
9154 v_gets = UInt64ToDouble(&vp->stats.hdr_gets);
9156 hdr_loads.sum += v_loads;
9157 hdr_gets.sum += v_gets;
9159 if (v_looks < ch_gets.min)
9160 ch_gets.min = v_looks;
9161 else if (v_looks > ch_gets.max)
9162 ch_gets.max = v_looks;
9164 if (v_loads < hdr_loads.min)
9165 hdr_loads.min = v_loads;
9166 else if (v_loads > hdr_loads.max)
9167 hdr_loads.max = v_loads;
9169 if (v_gets < hdr_gets.min)
9170 hdr_gets.min = v_gets;
9171 else if (v_gets > hdr_gets.max)
9172 hdr_gets.max = v_gets;
9175 /* compute per-chain averages */
9176 ch_gets.avg = ch_gets.sum / ((double)head->len);
9177 hdr_loads.avg = hdr_loads.sum / ((double)head->len);
9178 hdr_gets.avg = hdr_gets.sum / ((double)head->len);
9180 /* dump per-chain stats */
9181 Log("Volume hash chain %d : len=%d, looks=%s, reorders=%s\n",
9183 DoubleToPrintable(ch_looks.sum, pr_buf[0], sizeof(pr_buf[0])),
9184 DoubleToPrintable(ch_reorders.sum, pr_buf[1], sizeof(pr_buf[1])));
9185 Log("\tVolume gets : min=%s, max=%s, avg=%s, total=%s\n",
9186 DoubleToPrintable(ch_gets.min, pr_buf[0], sizeof(pr_buf[0])),
9187 DoubleToPrintable(ch_gets.max, pr_buf[1], sizeof(pr_buf[1])),
9188 DoubleToPrintable(ch_gets.avg, pr_buf[2], sizeof(pr_buf[2])),
9189 DoubleToPrintable(ch_gets.sum, pr_buf[3], sizeof(pr_buf[3])));
9190 Log("\tHDR gets : min=%s, max=%s, avg=%s, total=%s\n",
9191 DoubleToPrintable(hdr_gets.min, pr_buf[0], sizeof(pr_buf[0])),
9192 DoubleToPrintable(hdr_gets.max, pr_buf[1], sizeof(pr_buf[1])),
9193 DoubleToPrintable(hdr_gets.avg, pr_buf[2], sizeof(pr_buf[2])),
9194 DoubleToPrintable(hdr_gets.sum, pr_buf[3], sizeof(pr_buf[3])));
9195 Log("\tHDR loads : min=%s, max=%s, avg=%s, total=%s\n",
9196 DoubleToPrintable(hdr_loads.min, pr_buf[0], sizeof(pr_buf[0])),
9197 DoubleToPrintable(hdr_loads.max, pr_buf[1], sizeof(pr_buf[1])),
9198 DoubleToPrintable(hdr_loads.avg, pr_buf[2], sizeof(pr_buf[2])),
9199 DoubleToPrintable(hdr_loads.sum, pr_buf[3], sizeof(pr_buf[3])));
9200 } else if (flags & VOL_STATS_PER_CHAIN) {
9201 /* dump simple per-chain stats */
9202 Log("Volume hash chain %d : len=%d, looks=%s, gets=%s, reorders=%s\n",
9204 DoubleToPrintable(ch_looks.sum, pr_buf[0], sizeof(pr_buf[0])),
9205 DoubleToPrintable(ch_gets.sum, pr_buf[1], sizeof(pr_buf[1])),
9206 DoubleToPrintable(ch_reorders.sum, pr_buf[2], sizeof(pr_buf[2])));
9210 VHashEndExclusive_r(head);
9215 /* compute global averages */
9216 len.avg = len.sum / ((double)VolumeHashTable.Size);
9217 looks.avg = looks.sum / ((double)VolumeHashTable.Size);
9218 gets.avg = gets.sum / ((double)VolumeHashTable.Size);
9219 reorders.avg = reorders.sum / ((double)VolumeHashTable.Size);
9221 /* dump global stats */
9222 Log("Volume hash summary: %d buckets\n", VolumeHashTable.Size);
9223 Log(" chain length : min=%s, max=%s, avg=%s, total=%s\n",
9224 DoubleToPrintable(len.min, pr_buf[0], sizeof(pr_buf[0])),
9225 DoubleToPrintable(len.max, pr_buf[1], sizeof(pr_buf[1])),
9226 DoubleToPrintable(len.avg, pr_buf[2], sizeof(pr_buf[2])),
9227 DoubleToPrintable(len.sum, pr_buf[3], sizeof(pr_buf[3])));
9228 Log(" looks : min=%s, max=%s, avg=%s, total=%s\n",
9229 DoubleToPrintable(looks.min, pr_buf[0], sizeof(pr_buf[0])),
9230 DoubleToPrintable(looks.max, pr_buf[1], sizeof(pr_buf[1])),
9231 DoubleToPrintable(looks.avg, pr_buf[2], sizeof(pr_buf[2])),
9232 DoubleToPrintable(looks.sum, pr_buf[3], sizeof(pr_buf[3])));
9233 Log(" gets : min=%s, max=%s, avg=%s, total=%s\n",
9234 DoubleToPrintable(gets.min, pr_buf[0], sizeof(pr_buf[0])),
9235 DoubleToPrintable(gets.max, pr_buf[1], sizeof(pr_buf[1])),
9236 DoubleToPrintable(gets.avg, pr_buf[2], sizeof(pr_buf[2])),
9237 DoubleToPrintable(gets.sum, pr_buf[3], sizeof(pr_buf[3])));
9238 Log(" reorders : min=%s, max=%s, avg=%s, total=%s\n",
9239 DoubleToPrintable(reorders.min, pr_buf[0], sizeof(pr_buf[0])),
9240 DoubleToPrintable(reorders.max, pr_buf[1], sizeof(pr_buf[1])),
9241 DoubleToPrintable(reorders.avg, pr_buf[2], sizeof(pr_buf[2])),
9242 DoubleToPrintable(reorders.sum, pr_buf[3], sizeof(pr_buf[3])));
9244 /* print extended disk related statistics */
9246 struct DiskPartition64 * diskP;
9247 afs_uint32 vol_count[VOLMAXPARTS+1];
9248 byte part_exists[VOLMAXPARTS+1];
9252 memset(vol_count, 0, sizeof(vol_count));
9253 memset(part_exists, 0, sizeof(part_exists));
9257 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
9259 vol_count[id] = diskP->vol_list.len;
9260 part_exists[id] = 1;
9264 for (i = 0; i <= VOLMAXPARTS; i++) {
9265 if (part_exists[i]) {
9266 /* XXX while this is currently safe, it is a violation
9267 * of the VGetPartitionById_r interface contract. */
9268 diskP = VGetPartitionById_r(i, 0);
9270 Log("Partition %s has %d online volumes\n",
9271 VPartitionPath(diskP), diskP->vol_list.len);
9278 /* print extended VLRU statistics */
9279 if (VVLRUExtStats_r(&vlru_stats, vol_sum) == 0) {
9280 afs_uint32 idx, cur, lpos;
9285 Log("VLRU State Dump:\n\n");
9287 for (idx = VLRU_QUEUE_NEW; idx < VLRU_QUEUE_INVALID; idx++) {
9288 Log("\t%s:\n", vlru_idx_to_string(idx));
9291 for (cur = vlru_stats.queue_info[idx].start;
9292 cur < vlru_stats.queue_info[idx].len;
9294 line[lpos++] = vlru_stats.vec[cur].volid;
9296 Log("\t\t%u, %u, %u, %u, %u,\n",
9297 line[0], line[1], line[2], line[3], line[4]);
9306 Log("\t\t%u, %u, %u, %u, %u\n",
9307 line[0], line[1], line[2], line[3], line[4]);
9312 free(vlru_stats.vec);
9319 VPrintExtendedCacheStats(int flags)
9322 VPrintExtendedCacheStats_r(flags);
9325 #endif /* AFS_DEMAND_ATTACH_FS */
9328 VCanScheduleSalvage(void)
9330 return vol_opts.canScheduleSalvage;
9336 return vol_opts.canUseFSSYNC;
9340 VCanUseSALVSYNC(void)
9342 return vol_opts.canUseSALVSYNC;
9346 VCanUnsafeAttach(void)
9348 return vol_opts.unsafe_attach;